Calhoun: The NPS Institutional Archive

Theses and Dissertations Thesis Collection

2014-06

Cost Prediction via Quantitative Analysis of

Complexity in U.S. Navy Shipbuilding

Dobson, Aaron T.
Monterey, California. Naval Postgraduate School

http://hdl.handle.net/10945/43068
 COST PREDICTION VIA QUANTITATIVE ANALYSIS OF
COMPLEXITY IN U.S. NAVY SHIPBUILDING

by

Aaron T. Dobson
B.S., Aerospace Engineering, United States Naval Academy

Submitted to the Department of Mechanical Engineering and Engineering Systems

Division in Partial Fulfillment of the Requirements for the Degrees of

Naval Engineer
and
Master of Science in Engineering and Management
atthe
Massachusetts Institute of Technology
June 2014

© Aaron Dobson. All rights reserved.

The author hereby grants to MIT permission to reproduce and to
distribute publicly paper and electronic copies of this thesis document
in whole or in part in any medium now known or hereafter created.

Signature of Author:............................... . ...... . .... .. . .... ..

Department of Mechanical Engineering and~yste Design and Mana~ement
May 91 , 2014
.
Certified by: .... .. ............................ ............... . .. . .
Olivier L. de Weck
Professor of Aeronautics and Astronautics and Engineering Systems

Certified by: .................................................... :r".esis.S~p~~i~r

Eric S. Rebentisch

Certified b ~:o.~i.•:e: .s~c1~1~c~1~~-1. ~yste.~~~;~~~~~:~:~:

Professor of th . h...____

Accepted by: .......... . ......................

;t;;}cto~ Patrick Hale

~~~ _::.'.:~~:-~i~T;~
1

Accepted by:..............
David E. Hardt
Chairman, Department Committee on Graduate Studies
Department of Mechanical Engineering

1
 THIS PAGE INTENTIONALLY LEFT BLANK

! 2!
 Form Approved
REPORT DOCUMENTATION PAGE OMB No. 0704-0188

The public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data
sources, gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other
aspect of this collection of information, including suggestions for reducing the burden, to Department of Defense, Washington Headquarters Services, Directorate for Information
Operations and Reports (0704-0188), 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302. Respondents should be aware that notwithstanding any other
provision of law, no person shall be subject to any penalty for failing to comply with a collection of information if it does not display a currently valid OMB control number.
PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE ADDRESS.
1. REPORT DATE (DD-MM-YYYY) 2. REPORT TYPE 3. DATES COVERED (From - To)
Master's Thesis 15 May 2011-15 Jun 2014
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
COST PREDICTION VIA QUANTITATIVE ANALYSIS OF N00244-09-G0014
COMPLEXITY IN U.S. NAVY SHIPBUILDING 5b. GRANT NUMBER

5c. PROGRAM ELEMENT NUMBER

6. AUTHOR(S) 5d. PROJECT NUMBER

Aaron T. Dobson
5e. TASK NUMBER

5f. WORK UNIT NUMBER

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION

REPORT NUMBER
Massachusetts Institute of Technology

9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR'S ACRONYM(S)

Naval Postgraduate School NPS
Monterey, CA 93943
11. SPONSOR/MONITOR'S REPORT
NUMBER(S)

12. DISTRIBUTION/AVAILABILITY STATEMENT

Approved for public release; distribution is unlimited.

13. SUPPLEMENTARY NOTES

14. ABSTRACT
The primary objective of this study was to quantify, assess, and analyze cost-predictive complexity-oriented benchmarks in
the pre-construction phase of the U.S. Navy’s ship acquisition process.
The end result was that a consideration of complexity via the methods and algorithms established in this study supported an
exponential cost versus complexity relationship to refine the current cost estimation methods and software
currently in use in U.S. Navy shipbuilding.

15. SUBJECT TERMS

16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF 18. NUMBER 19a. NAME OF RESPONSIBLE PERSON
ABSTRACT OF
a. REPORT b. ABSTRACT c. THIS PAGE Carol Stoker
PAGES
19b. TELEPHONE NUMBER (Include area code)
UU 111 (831) 656-3025
Standard Form 298 (Rev. 8/98)
Prescribed by ANSI Std. Z39.18
 COST PREDICTION VIA QUANTITATIVE ANALYSIS OF
COMPLEXITY IN U.S. NAVY SHIPBUILDING

by

Aaron T. Dobson

Submitted to the Department of Mechanical Engineering and Engineering Systems

Division on May 9th, 2014 in Partial Fulfillment of the Requirements for the Degrees of
Naval Engineer and Master of Science in Engineering and Management

Abstract
As the sophistication and technology of ships increases, U.S. Navy shipbuilding must be
an effective and cost-efficient acquirer of technology-dense one-of-a-kind ships all while
meeting significant cost and schedule constraints in a fluctuating demand environment.
A drive to provide world-class technology to the U.S. Navy’s warfighters necessitates
increasingly complex ships, which further augments the non-trivial problem of providing
cost effective, on-schedule ships for the American taxpayer. The primary objective of
this study was to quantify, assess, and analyze cost-predictive complexity-oriented
benchmarks in the pre-construction phase of the U.S. Navy’s ship acquisition process.
This study used commercially-available software such as Mathwork’s MATLAB
software to analyze the numerical cost data and assess the fidelity of the predictive
benchmarks to the datasets. The end result was that a consideration of complexity via the
methods and algorithms established in this study supported an exponential cost versus
complexity relationship to refine the current cost estimation methods and software
currently in use in U.S. Navy shipbuilding. Specifically, it was found that for the
subsystems under analysis, acquisition/contract cost per unit was highly correlated with
unit complexity according to the relationship, cost/unit ($M,USD) = 23.100 + e0.015C.

Thesis Supervisor: Olivier L. de Weck

Title: Associate Professor of Aeronautics and Astronautics and Engineering Systems

Thesis Supervisor: Eric S. Rebentisch

Title: Research Associate, Production in the Innovation Economy Study

Thesis Supervisor: Mark W. Thomas

Title: Professor of the Practice of Naval Construction and Engineering
 THIS PAGE INTENTIONALLY LEFT BLANK

! 4!
Table of Contents
ABSTRACT 3

LIST OF FIGURES 6

LIST OF TABLES 7

LIST OF ACRONYMS AND ABBREVIATIONS 8

BIOGRAPHICAL NOTE 9

ACKNOWLEDGEMENTS 10

1. PROBLEM BACKGROUND AND THE DDG 51 CASE STUDY 11

1.1 DDG 51 ARLEIGH BURKE-CLASS GUIDED MISSILE DESTROYERS: A CASE STUDY 13
1.2 PRODUCTION IN THE INNOVATION ECONOMY (PIE) STUDY 15

2. LITERATURE REVIEW 17
2.1 COST GROWTH 18
2.2 STRUCTURAL COMPLEXITY 20

3. INFORMATION AND DATA COLLECTION 28

3.1 DATA COLLECTION: INTERVIEWS 29
3.2 DATA COLLECTION 29
3.2.1 COMPONENT COMPLEXITY METRIC, XI 30
3.2.2 INTERFACE COMPLEXITY ASSESSMENT, Β 33
3.2.3 FUNCTIONAL BLOCK DIAGRAMS AND SCHEMATICS 35

4. SHIP SUBSYSTEM COMPLEXITY 47

4.1 AMDR/AEGIS, MCS, AND MRG 47
4.2 COST ASSESSMENT 50
4.2.1 MCS COST ASSESSMENT 51
4.2.2 MRG COST ASSESSMENT 54
4.2.3 AEGIS AND AMDR COST ASSESSMENT 55

5. RESULTS AND ANALYSIS 58

5.1 AGGREGATED SUBSYSTEM COMPARISON AND ANALYSIS 58
5.2 SENSITIVITY ANALYSIS 61
5.3 DISCUSSION 66

6. SUMMARY AND RECOMMENDATIONS 69

6.1 SUMMARY 69
6.2 RECOMMENDATIONS 71
6.3 AREAS FOR FURTHER STUDY 72

! 5!
6.4 CLOSING COMMENTS 73

BIBLIOGRAPHY 76

APPENDICES 81
APPENDIX A: TRAVEL SUMMARY AND INTERVIEW WITH TECHNICAL DIRECTOR, PMS 400D
81
APPENDIX B: MATLAB COMPLEXITY ALGORITHM, SOURCE SCRIPT, AND OUTPUT 85
APPENDIX C: U.S. SHIPYARDS 91
C.1 U.S. NEW CONSTRUCTION SHIPBUILDERS AND SHIPYARDS 91
C.2 U.S. REPAIR, MODERNIZATION, AND OVERHAUL (RMO) SHIPYARDS 95
APPENDIX D: R.O.K. SHIPYARDS AND THE KDX-CLASS 97
D.1 REPUBLIC OF KOREA (R.O.K.)’S SHIPYARDS 97
D.2 R.O.K.’S KDX PROGRAM 98
APPENDIX E: BENCHMARKING IN NAVAL SHIPBUILDING 101
!

List of Figures
!
Figure 1: DDG 51 Arleigh Burke Guided Missile Destroyer 14
Figure 2: DDG 51 Class Evolution by Flight 15
Figure 3: Cost Growth in US Navy Warships 19
Figure 4: Topological Complexity 24
Figure 5: Matrix Energy Nodal Structure Example 27
Figure 6: Research, Constructs, and Data Flow 28
Figure 7: TRL Level Definitions and Expanded Definitions 31
Figure 8: Sensitivity Analysis for AMDR K-Factor 38
Figure 9: AEGIS Subsystem - Unclassified 39
Figure 10: Machinery Control System - Unclassified 42
Figure 11: DDG 106 Propulsion Reduction Gear Arrangement, Port Gear, Isometric View -
Unclassified 45
Figure 12: Complexity Component Breakout 49
Figure 13: DoD Cost Types and Relationships 50
Figure 14: Machinery Control System Learning Curve Slopes 53
Figure 15: RCA's 1969 Contract Initiation 55
Figure 16: Cost versus Complexity and Cost Variance 59
Figure 17: C3 versus Percentage Increase in Subsystem Interconnectivity 64
Figure 18: DOD UAV Roadmap 66
!.!"#$
Figure 19: Unit Cost vs. Complexity, Exponential Fit: $M = 23.1e 70
Figure 20: Tier One U.S. Shipbuilders and Their Subsidiaries 92
Figure 21: GD Revenues by Shipyard and Sector 93
Figure 22: U.S. Navy New Construction Apportionment by Revenue 94
Figure 23: Worldwide Shipbuilding Market Share 97
Figure 24: Sejong the Great (DDH 991) and DDG 80, a DDG 51 class Flight IIA vessel 99
Figure 25: Analytical Process Underlying GSIBBS Findings 102
Figure 26: Notional Benchmarking Results 104
Figure 27: Comparison of Vessel Work Content by the CGT Method 105
Figure 28: Trends in Productivity vs. Use of Best Practice 107
Figure 29: Productivity vs. Overall Best Practice Rating 108

! 6!
Figure 30: US and International Benchmarking 109
Figure 31: Significant Factors Undermining U.S. Shipyard Core Productivity 109

List of Tables
Table 1: DDG 51 Class Evolution by Flight 15
Table 2: Dimensions of Project Complexity 17
Table 3: Subsystem Risk and X6 Values 32
Table 4: X-Vector Summary 33
Table 5: Subsystem Interface Complexity, βij 35
Table 6: Adjacency Matrix, AAEGIS 40
Table 7: Binary Adjacency Matrix, AMCS 43
Table 8: Binary Adjacency Matrix, AMRG 46
Table 9: Complexity Component Results 48
Table 10: Typical Slope Values by Activity 52
Table 11: AEGIS Upgrade Series Program Cost Estimation 56
Table 12: Subsystem Cost Summary, in Millions USD 58
Table 13: Notional MCS Adjacency Matrix with 100% Increase in Interconnectivity 63
Table 14: Tier One, Tier Two, and Public Shipyards vs. Ship Classes and Respective
Functional Alignment 96
Table 15: KDX Program Summary (Naval Technology 2013) 100!

! 7!
List of Acronyms and Abbreviations
AMDR Air and Missile Defense Radar
ASN Assistant Secretary of the Navy
CBO Congressional Budget Office
CGT Compensated Gross Tonnage
CVN Aircraft Carrier, Nuclear
DDG Destroyer, Guided Missile
EVM Earned Value Management
EVMS Earned Value Management System
FFG Frigate, Guided Missile
FFP Firm Fixed Price
FMI First Marine International
FY Fiscal Year
GAO Government Accountability Office
GD General Dynamics
GSIBBS Global Shipbuilding Industrial Base Benchmarking Study
GFE Government Furnished Equipment
HII Huntington Ingalls Industries
HM&E Hull, Mechanical, and Electrical
JHSV Joint High Speed Vessel
KDX Korean Destroyer Experimental
LCS Littoral Combat Ship
LHA Landing Helicopter Assault
LPD Landing Platform Dock
LSD Landing Ship Dock
MIT Massachusetts Institute of Technology
MCS Mission Control System
MLP Mobile Landing Platform
MRG Main Reduction Gear
NAVSEA Naval Seas Systems Command
NASSCO National Steel and Shipbuilding Company
NG Northrop Grumman
NGSB Northrop Grumman Shipbuilding
NNSY Norfolk Naval Shipyard
PEO Program Executive Office
PHNSY Pearl Harbor Naval Shipyard
PGC Philadelphia Gear Company
PIE Production in the Innovation Economy
PNSY Portsmouth Naval Shipyard
PMS Program Management Office (Ships)
PSNSY Puget Sound Naval Shipyard
RD&A Research, Development, and Acquisition
ROK Republic of Korea

! 8!
RMO Repair, Modernization, and Overhaul
SOW Statement of Work
USNS United States Naval Ship (Non-commissioned)
USN United States Navy
USS United States Ship (Commissioned)
T-AKE Auxiliary Cargo (K) and Ammunition (E) Ship
ZEDS Zonal Electrical Distribution System

Biographical Note

Lieutenant Aaron Dobson is a native of Southlake, Texas, and he graduated from the US
Naval Academy in 2005 with a Bachelor of Science in Aerospace Engineering. After
graduation and commissioning, he attended Navy Pilot training in Corpus Christi, Texas,
where he earned his wings in March 2007 and then went on to fly the P-3C Orion.

In the spring of 2008, LT Dobson made a direct accession to the Engineering Duty
Officer Community and reported for his qualification tour at SPAWAR Space Field
Activity in Chantilly, Virginia. After serving in two communication satellite acquisition
programs, LT Dobson reported to Massachusetts Institute of Technology (MIT) in April
2011. At MIT Dobson aspires to earn his Engineer’s Degree in Naval Architecture &
Marine Engineering and a Master of Science in Engineering & Management. He will
graduate in June 2014, and report to Sasebo, Kyushu, Japan for duty as Ship
Superintendent of various ship classes.

Dobson’s future goals are to combine the business education and naval architecture
education received at MIT for application in the field of surface ship program
management.

! 9!
Acknowledgements

Captain Mark Thomas, Professor Olivier de Weck, and Dr. Eric Rebentisch’s persistent
and insightful guidance made this thesis possible. Without their mentorship, very little of
this research would have been possible.

Fred Harris, Tom Wetherald, Duke Vuong, and the entire team at General Dynamics
NASSCO in San Diego, California for their outstanding support of the MIT team’s visit
to their facility on August 13th, 2013. Mr. Harris has gone above and beyond in support of
our research and in the pursuit of the betterment of Navy and American shipbuilding as a
whole.

Todd Hellman at Naval Sea Systems Command PMS 400D provided much needed and
valuable information in the furtherance of this thesis.

Finally, I want to thank my beautiful wife Amanda for her enduring support, love, and
perceptive insights.

! 10!
1. Problem Background and the DDG 51 Case Study

Over the past 50 years, the cost of U.S. Naval Shipbuilding has grown between 7 to 11%
per year, far outpacing inflationary effects during the same timeframe (RAND
Corporation 2006). Although the Navy has migrated away from purely weight-based cost
estimation methods in the last decade, unpredictable cost growth remains an issue that
costs the U.S. taxpayers billions of dollars annually. Background research revealed that
cost growth has arisen from two main sources: customer-driven factors and economy-
driven factors, both of which will be explored in depth in Section 2. It was assumed that
economy-driven factors in cost growth are of a sufficiently “macro” level to be beyond
the reach of Navy policy makers, program managers, and cost estimators to manage or
predict. Therefore, the focus narrowed to what was within the Navy’s grasp to control:
the customer-driven factors, with the Navy functioning as the acquiring agent of ships.

It was discovered that reports on cost growth in U.S. Navy shipbuilding repeatedly
returned to the theme of the ever-increasing complexity in modern Naval vessels as a
substantial contributor to cost growth. While technological advancement has taken the
Navy from relatively simple cannons to highly sophisticated missiles able to obliterate
satellites in Low Earth Orbit in the span of a century, the effects on cost growth to build
those subsystems has been substantial.

In order to understand and gain traction on the concept of complexity in Naval

shipbuilding, it was necessary to determine a method with which to quantify that
complexity. Fortunately, MIT doctoral student (at the time of writing) Kaushik Sinha and
MIT Professor Olivier de Weck developed an algorithm that did precisely that: quantified
structural complexity in a mostly generalizable manner, and although their research
focused primarily on software-intensive hardware systems, the equations, algorithms, and
logic were modified to suit an application to Naval maritime systems.

! 11!
As with any algorithm, the quality of the outputs is only as good as the quality of its
inputs, so it was determined that a suitable case study needed to meet several
requirements:
• Class longevity. A long history of actual (vice predicted, parametric, or
analogous) costs would help reliably determine what, if any, relation complexity
had to cost.
• Proliferation of ships in class. A class with a large number of ships in class would
yield more data than a smaller class. It was also hypothesized that any benefits
gained from this analysis would positively affect more ships in a larger class.
• Ship class currently in production. There was a desire to choose a class that had
not terminated its production run so as to garner at least one or two data points to
serve as a predictive measure for the algorithm. The goal for this research was not
to merely tell the reader what was, but what could be in terms of cost growth
potential in major subsystems.

Given the diverse set of requirements imposed upon the given ship classes, it was
determined that the U.S. DDG 51 Arleigh Burke-class and its sister class, the Republic of
Korea’s KDX Sejong the Great-class would be suitable classes of ship for a case study.
Both classes have witnessed a production run since the 1980s, with ships populating
either three or four individual Flights within class, and are still currently under production
in the host country. Although the original desire was to study both ships in their
respective shipyards, the R.O.K.’s security concerns precluded the in-depth study of the
KDX-class so this led to focusing this study on the system and subsystems within the
DDG 51 vessels.

To summarize, the main focus of the research in this study was to examine the concept of
the characteristic complexity inherent in various DDG 51 subsystems that drive cost
growth and cost uncertainty. It is hypothesized that direct correlations between cost and
complexity could drive down cost uncertainty for U.S. Navy policymakers, save the
taxpayer large sums of money, and help refine current cost-predictive software tools
currently in use by the Navy’s cost estimation groups.

! 12!
1.1 DDG 51 Arleigh Burke-class Guided Missile Destroyers: A Case Study

Since the launch of the original DDG 51, the U.S.S. Arleigh Burke, in 1985, variants of
the DDG 51 concept have garnered international popularity as highly capable, multi-
mission platforms featuring the powerful AEGIS radar system (FAS Military Network
2013). While the U.S. concept will be discussed in further detail later, several other
countries such as Japan, Spain, Norway, and Australia either have or will have employed
variations of AEGIS-capable destroyers. A brief discussion of the aforementioned KDX-
class and the Korean shipyard environment is included in Appendix D.

Selecting a vessel of comparable complexity between the U.S. shipyards and the R.O.K.
shipyards facilitates a more relevant benchmarking study by comparing “apples to
apples” versus comparing a shipyard producing relatively high complexity vessels such
as naval combatants versus a shipyard producing relatively low complexity vessels such
as container ships.

The DDG 51 class is the U.S.’s “jack of all trades” guided missile destroyer, and while it
was originally designed to combat and defend against Soviet-era threats, the employment
of the highly sophisticated AEGIS air defense system has allowed the craft to evolve into
several different modern mission areas. These combat capabilities include anti-air, anti-
submarine, anti-surface, strike operations, and ballistic missile defense (FAS Military
Network 2013). A representative multi-view of the ship is shown in Figure!1.

! 13!
 !
Figure 1: DDG 51 Arleigh Burke Guided Missile Destroyer (USS DDG 51 Arleigh
Burke Destroyer 2009)

At the time of writing, three flights of the DDG 51 class are currently at sea, and
requirement evolution is underway on a fourth flight. In June 2013, the Navy announced
that two contracts were awarded to Bath Iron Works (BIW) and Huntington Ingalls
Industries (HII) for continued construction of the Flight IIA and eventual construction of
Flight III currently under requirements development. As shown in Table!1 and Figure!2,
Flight III is expected to begin construction in FY 2016 and will likely provide increased
capability via an increase in power and cooling to accommodate replacing the AEGIS
AN/AEGIS radar with the Air and Missile Defense Radar (AMDR). (NAVSEA Office of
Corporate Communication 2013) (Captain Mark Vandroff, Program Manager, DDG 51
Shipbuilding Program 2012)

The DDG 51 class evolution is shown in Figure 2 where hull numbers starting with the
original Arleigh Burke and progressing up through the proposed Flight III vessels are
mapped to their respective fiscal years and builders.

! 14!
 Figure 2: DDG 51 Class Evolution by Flight (Captain Mark Vandroff, DDG 51
Program 2013)

Table 1: DDG 51 Class Evolution by Flight. Numbers in italics represent projected

figures based on current data. (NAVSEA Shipbuilding Support Office 2013)

Flight Number Hulls Launch Periods Lightship Weight

I 51 – 71 1989 - 1996 6691 – 6827 tons
II 72 – 78 1996 - 1997 6805 – 6824 tons
IIA 79 – 122 1998 - 2015 7134 – 7134 tons
III 123+ 2016+ No Data

1.2 Production in the Innovation Economy (PIE) Study

In 2011, the Massachusetts Institute of Technology (MIT) commissioned the PIE study,
and its “overarching goal is to shed light on how America's great strengths in innovation
can be scaled up into new productive capabilities. The goal is to develop

! 15!
recommendations for transforming America's production capabilities in an era of
increased global competition.” (MIT 2011) The Assistant Secretary of the Navy for
Research, Development, and Acquisition (ASN(RD&A)) further funded the study to add
a standalone module to address US shipbuilding.

There are five parts in the Technical Statement of Work (SOW) of the shipbuilding
module of the PIE study.
1. Innovation in Bidding and Contracting
2. Project Management and Rework Dynamics
3. National and International Benchmarking of U.S. Shipbuilding Performance
4. Supply Chain Management and Supplier Base
5. Prospects for U.S. Commercial Shipbuilding

This thesis research was conducted in part for SOW task two and three.

On directives received from the ASN(RD&A)’s office, “the scope of the study is the
shipbuilding industry, holding other stakeholders as part of the boundary conditions. The
scope includes the linkage between the contractor and USN through the program
lifecycle.” The study seeks to answer two overarching questions:
• Can the U.S. government be doing more to put more pressure on the prime
contractors and suppliers to get better cost performance and innovation?
• Furthermore, why is the escalation in shipbuilding costs greater than general
inflation?

! 16!
2. Literature Review

The literature review focused on two main topics: first, a review of the cost growth
problem within the field of Naval shipbuilding, and secondly, an examination of Kaushik
Sinha and Olivier de Weck’s research into structural complexity quantification will be
reviewed for later application to ships and shipboard systems.

The overall topic of complexity was broken into three dimensions: technical,
organizational, and strategic complexity (Hoffman and Kohut 2012) as shown in Table!2.

Table 2: Dimensions of Project Complexity (Hoffman and Kohut 2012)

Technical Organizational Strategic

Number and variety of partners Number and
Number and type of interfaces (industry, international, diversity of
academia/research) stakeholders
Technology development Distributed/virtual team; Socio-political
Requirements decentralized authority context
Interdependencies among
Funding sources
technologies (tight coupling Horizontal project organization
and processes
vs. loose)

For complex, technical projects fully understanding the interfaces and interdependencies
in a given system or subsystem are crucial to success. George Low, the legendary leader
of NASA’s Apollo program, knew this was a key to Apollo’s success. He noted that only
100 wires linked the Saturn rocket to the Apollo spacecraft. “The main point is that a
single man can fully understand this interface and can cope with all the effects of a
change on either side of the interface. If there had been 10 times as many wires, it
probably would have taken a hundred (or a thousand?) times as many people to handle
the interface,” he wrote.” (Hoffman and Kohut 2012) A similar interface philosophy
applied to Navy projects such as Littoral Combat Ship could not only yield a decrease in
cost, but also an increase in operational tempo due to decreased time in port between
mission module swaps.

! 17!
Methods presented in Section 2.2 and applied in Section 4 focused on the technical
dimension of complexity, but as Table!2 implies, the concept of complexity has a far
greater diversity of roots than simply the technical aspect alone. While the algorithms and
quantification of complexity focused on the technical aspect, Section 5 attempted to unify
these dimensions, with the principal results from the algorithm logic, into actionable
points for cost savings for the U.S. taxpayer.

2.1 Cost Growth

!
Over the past 50 years, the cost of US naval shipbuilding has increased between 7 to 11%
annually, while the average inflation from 1913 to 2013 has been approximately 3.22%
(Inflation Data 2013). Given the current environment of fiscal austerity imposed upon
government spending, the ever-increasing cost of has ships has resulted in subsequently
squeezed naval shipbuilding budgets. A 2006 RAND Corporation study based on
Congressional Budget Office (CBO) data points out that a hypothetical boost of $2
billion to $12 billion dollars would “only help the Navy achieve a fleet of 260 ships by
the year 2035 rather than the nearly 290 it now has.” (RAND Corporation 2006) At the
time of writing in 2013, the current US Navy fleet size of commissioned ships has been
reduced to 250 ships (NAVSEA Shipbuilding Support Office 2013).

While many factors contribute to ship cost growth, those factors can be attributed to two
broad categories: economy-driven factors and customer-driven factors. Economy-driven
factors include items such as equipment, labor rates, and material costs, and the 2006
RAND study found that the cost growth in those rates tracked closely with overall U.S.
inflation rates. These economy-driven factors were responsible for a 3.41%1 cost increase
between fiscal years 1961 and 2002 (RAND Corporation 2006), which leaves the
remaining 2.5 to 6.5% of cost growth attributable to customer-driven factors.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
1
3.41% was the average inflation in the U.S. between 1960 and 2009 (Inflation Data
2013).

! 18!
Customer-driven cost increases were grouped into three subcategories with their
associated percentage cost increase between DD 2 in FY 61 and DDG 51 in FY 2002
(RAND Corporation 2006):
• Characteristic Complexity (2.1%)
• Standards, Regulations, and Requirements (2.0%)
• Procurement Rates (0.3%)

The magnitude of cost growth for different ships is shown in Figure!1 from a 2005 GAO
report.

Figure 3: Cost Growth in US Navy Warships (GAO 2005)

Although the 2006 RAND study was the only report discussed in this paper, several GAO
reports, program briefings, and third-party maritime consultants have cited complexity as
being a primary driver in cost growth. The end result of the cost growth research
determined that there is an inextricable link between complexity and cost growth, and the
fact that this complexity is a customer-driven factor warrants a discussion to answer the
questions of how do we as the U.S. Navy gain traction and understanding on the concept
of complexity and refine our cost estimation techniques to capture the notion of
complexity? What complexity-based mitigation factors, if any, are necessary to drive

! 19!
down cost growth and better assess what the final price tag would be? This research
sought to not to reduce the cost of shipbuilding itself, but to gain a better definitive
understanding of how complexity drives cost in an effort to shrink the gap between
projected and actual costs.

2.2 Structural Complexity

To gain a hold on the concept of complexity and provide a quantitative baseline with
which to compare different systems, this research focused on work published by Sinha
and de Weck, 2013 at MIT. This work served as the numerical basis for the analysis of
the DDG 51 class, which will occur in Section 4. The method for quantifying structural
complexity was selected as a basis for analysis due to its generalizability of application in
engineering systems. Sinha and de Weck, 2013 validated their method for a large (and
thus generalizable) range of systems including those of high complexity such as a
satellite, aircraft engines, and those systems of low complexity, such as a hair dryer.

Structural complexity was described in functional form via the following simple relation:

Structural Complexity, C = C1 + C2C3

• C1 represented the “sum of complexities of individual components alone (local

effect) and does not involve architectural information.” (Sinha and de Weck
2013). This term was directly associated with activities related to component
engineering. If a system was completely disassembled and all the components
(hardware and software) spread out on a hypothetical flat plane, this term would
represent the sum of all the individual component complexities.

To capture the multi-faceted nature of component complexity C1, Sinha and de

Weck sought to aggregate the factors in a way where each particular dimension of
C1 could be analyzed, quantified, and then subsequently rolled up into the parent
C1 figure using an algorithm that normalizes the terms and then sums them. This
initial aggregation of factors is defined in an n x 1 array called the X-vector. Sinha

! 20!
 and de Weck, 2013 initially proposed the terms within the X-vector (Sinha and de
Weck 2013) as shown below, but that vector has been adapted for this DDG-51
class-oriented case study. In Sinha and de Weck’s original work, C1(X(i)) was
primarily adapted for applications in Cyber-Physical Systems and thus several
terms were substituted for more maritime-suitable counterparts.

(!) (!)
!! !! = !(!"#$%#&'()"!!"#$%&'($)
(!) (!)
!! !! = !(!"#$%#&'()"!!"#"$)
(!) (!)
!! !! = ! !"#$"%&%'!!"#$
(!) (!)
(!)
!! !! = !(!"#$%&!!"!!!"#$%&!!"#$"%&"'(#)
! = (!)
= (!)
!! !! = !(!"#$%&!!"!!"#$"%&'()
(!) (!)
!! !! = !(!"#)
(!) (!)
!! !! = !(!"#$%&'"!!"#$%&'(&)
(!) (!)
!! !! = !(!"#$%&'"!!"#$")

For!the!purposes!of!this!research,!the!X?vector!is!defined!as!follows:!
• X1!–!Measure!of!performance!tolerance.!Systems!and!subsystems!with!
smaller!tolerances!in!performance!garner!a!higher!X1!rating!than!those!that!
do!not.!For!example,!a!missile!tasked!to!strike!a!1!square!meter!target!would!
have!a!higher!value!for!X1!than!another!missile!tasked!to!strike!a!10!square!
meter!target.!
• X2!–!Performance!level.!What!is!the!performance!level!expected!of!the!system!
or!subsystem?!!It!is!posed!that!performance!correlates!with!complexity!in!the!
same!way!that!an!Italian!supercar!is!more!complex!than!an!old!pickup!truck.!!!!
• X3!–!Component!“size”!indicator.!How!big!is!the!system?!!X3!captures!the!
complexity!of!size;!for!example,!it!is!proposed!that!an!office!building!is!more!
complex!than!an!average!single!residential!house.!!!
• X4!–!Number!of!coupled!disciplines!involved.!How!many!different!fields!are!
coming!together!to!produce!this!product?!A!purely!mechanical!interface!is!
most!likely!less!complex!than!an!electrical!to!mechanical!interface,!and!X4!

! 21!
 captures!the!degree!to!which!varied!disciplines!must!come!together!in!a!
product.!
• X5!–!Number!of!variables!involved.!
• X6!–!Reliability.!The!concept!of!reliability!in!the!X6!factor!has!been!adapted!in!
this!study!via!the!use!of!DoD’s!Technology!Readiness!Level.!X6!is!a!function!of!
the!top!of!the!TRL!scale!divided!by!the!unit’s!actual!assessed!TRL.!TRL!and!
the!DoD!definition!of!TRL!will!be!discussed!in!further!detail!below.!
!
For US systems acquired by the Department of Defense, relative technological
development is quantified through the use of the Technology Readiness Level
(TRL) System. Fielded, mature systems are rated at a 9, and technologies that
only exist in theory and rely on new concepts from science are rated at a TRL of
1. The TRL spans discretely (vice a continuum) from 1 to 9 because of the
nonlinear financial and schedule impacts not captured by a purely linear
continuum.

!"#!"#
!! =
!"#!

The concept of TRL and each item’s specific relation to x6 will be explained in
greater detail later.
!
• X7!–!Existing!knowledge!of!operating!principle.!!This!metric!examines!to!what!
degree!the!technology!under!question!is!“new”!technology.!Is!the!component!
under!analysis!a!product!of!old!and!theoretically!well?understood!knowledge!
or!is!the!subsystem!under!question!relatively!novel?!As!understanding!
increases,!X7!decreases.!For!example,!the!mechanical!workings!of!a!bicycle!
could!appear!quite!complex!to!the!uninitiated,!but!to!a!veteran!racer!or!
bicycle!mechanic!the!bicycle!is!a!comparatively!simple!machine!due!to!their!
degree!of!existing!or!prior!knowledge!of!the!principles!involved.!!!

! 22!
 • X8!–!Extent!of!reuse/heritage!indicator.!Similarly!to!X7,!an!increase!in!
heritage!reuse,!would!signify!a!decrease!in!X8!for!similar!reasons.!!!
!
C1!was!the!summation!of!its!parts!(in!this!case,!the!X?vector),!scaled!for!
commonality!to!a!range!between!0!and!10.!Although!it!is!possible!to!weight!the!
component!terms!of!C1!such!that!some!terms!are!more!important!than!others,!there!
did!not!appear!to!be!an!objective!initial!reason!to!do!so.!!!

!!!

!! = ! !!
!!!

• The second term C2 represents the number and complexity of each pair-wise
interaction. C2 comprises the “interfaces” term and is related to the design and
management of interfaces between the individual components described in C1.

To calculate C2 each α components’ pairwise interface is defined by a certain β

value comprised of two nonzero α values and their coefficient characteristic for
that type of interface. Two complex components that interface directly are more
likely to have a complex interface compared to a single complex component
interfacing with a simple component or two simple components interfacing with
each other directly.

!!" = !!" !! !! !!!ℎ!"!!!! , !! ! ≠ 0

Again, for n components and m interactions,

! !

!! = ! !!" !!"
!!! !!!

! 23!
 • C3 represents the global effect produced by the effect of architecture or
arrangement on the interfaces in a specific topology. This topological complexity
term is based upon the product’s architecture and is related to the required System
Integration efforts.

One of the key contributors to complexity is the internal architectural complexity

of the system, and on the physical level that complexity is manifested through the
interconnectedness of the system or subsystem. That internal complexity is
captured through the use of the Design Structure Matrix (DSM). Figure!4
represents three structural examples of increasing topological complexity. From
left to right, these structures represent a centralized/bus architecture, a hierarchical
architecture, and a distributed architecture. The degree of interconnectedness of
the nodes is the chief differentiating factor leading to increased complexity in the
C3 variable.

Figure 4: Topological Complexity

Measurement of the topological complexity in the ship’s identified subsystem is

captured via matrix energy. A property of matrix energy is that it is invariant to
isomorphic transformations of the matrix, which makes it a suitable method for
measuring the interconnectedness of the subsystem (Horn and Johnson 1994). The
final topological complexity metric is defined by the energy of that adjacency
matrix A ∈ Mnxn.

! 24!
 The concept of matrix (or graph) energy was introduced by Ivan Gutman in 1978
as an aid for modeling the π-electron energy of molecules. Gutman “formulated
the π-electron energy of certain molecules as the sum of the absolute values of the
eigenvalues of the adjacency matrix of a chemical graph associated with the
molecular bonds.” (diStefano and Davis 2009)

In each identified subsystem, Λ is the set of connected nodes in the DSM Aij.
(Sinha and de Weck 2013)

1, ∀[(!, !)| ! ≠ ! !and! !, ! ∈ !]

!!" =
0, !!"ℎ!"#$%!

The energy is given as follows based on Gutman’s method and the standard
eigenvalue equation (diStefano and Davis 2009):

! !!" = !! ; !!!! − !!! = 0!!

!!!

Unidirectional interfaces only.

One of the chief limitations of the relatively simple summation of the absolute
value of the eigenvalues is its lack of generalizability in undirected interfaces. In
engineered systems, an undirected interface can be thought of as one in which
data can flow multiple directions. For a physical system such as a welded joint,
the system is directed, and with that system, the summation of eigenvalues would
be sufficient, but in order to make this analysis as generalizable as possible,
especially in regards to the analysis of advanced sensors and weaponry, the
summation of singular values via a singular value decomposition will be used in
the calculation of C3.

! 25!
 !

! !!" = !! ; !!!! − !!! = 0

!!!

Graph energy for Adjacency matrix Aij, containing a binary representation of

generalizable, omnidirectional/undirected interfaces.

For the examples given below, A1 would have an adjacency matrix represented by
the following relatively sparse matrix given that only the central node has
connections to the others, while A2 has a slightly higher degree of connectivity
among the nodes:

0 1 1 1 1 1 1 0 1 1 0 0 0 0
1 0 0 0 0 0 0 1 0 0 1 1 0 0
1 ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ 1 0 ⋮ 0 0 1 1
!! = 1 ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ;!!!! = 0 1 ⋮ ⋮ ⋮ ⋮ ⋮
1 ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ 1 0 ⋮ ⋮ ⋮ ⋮
1 ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ 0 1 ⋮ ⋮ ⋮ ⋮
1 0 0 0 0 0 0 0 0 1 0 0 0 0

From this equation one can calculate that λ1 = -2.45 and λ2 = 2.45 (all other
eigenvalues are 0) so E(A1) = 4.9. Following the same method, one can calculate
the notional E(A2) value is 6.83 (λ1 = -2, λ2 = 2, λ3 = -1.41, λ4 = 1.41, λ5-7 = 0)
which is indicative of A2 being the more topographically complex structure by its
exhibition of a higher degree of inter-connectedness as shown in Figure!5, even
though both structures A1 and A2 have the same number of nodes and edges,
respectively.

! 26!
 Figure 5: Matrix Energy Nodal Structure Example

!
! ! = !!! !! , where σι represents the ith singular value

C3 was equal to the matrix energy expressed from above times a normalization factor γ
based on the number of components.

C3 = γ E(A), γ = 1/n

In summary, structural complexity can be quantified via the integration of the discussed
terms into the original equation for each subsystem targeted for analysis.

! ! !

! !, !, ! = ! !! + ! !!" !!" !!(!)

!!! !!! !!!

As an illustrative example, Sinha and de Weck, 2013 showed that this analysis applied to
two different jet engine architectures, “namely a dual spool direct-drive turbofan (e.g.
new architecture) and a geared turbofan engine (e.g., new architecture)”. After consulting
with experts to collect data regarding interface complexity, it was determined structural
complexity was underestimated by 43% since originally only the amount of connections
and pair-wise interfaces were considered. This resulted in a real 40% increase in
complexity when only 28% was predicted thus contributing to an increase in
development cost over the previous turbofan engine’s predecessor.

! 27!
3. Information and Data Collection

All information and data collection efforts were seeking to answer the two main
constructs that comprise the main premise of this study:
1. Complexity: The number of elements and interdependencies in the ship’s
architecture contributes to the ship’s overall complexity.
2. Technical Risk: As a program management tool, technical risk is a key indicator
of a program’s self-assessed vulnerabilities. Those areas typically lie in areas of
high complexity where the technology or employment of that technology is least
understood. For example, a program with a requirement to employ a new radar
will most likely assess that radar at a higher technical risk level than another
shipboard system such as the hull.
This information is summarized in Figure!6.

Figure 6: Research, Constructs, and Data Flow

! 28!
3.1 Data Collection: Interviews

Interviews were conducted throughout the research at several different venues: organized
shipyard tours, conferences, and appointments. A summary of the interviews, meetings,
and travel conducted in support of this research is included in Appendix A. As the
research progressed, the nature of the interviews evolved from seeking contextual and
background information to seeking pertinent data and best practice information.

At the beginning of the research, the interviews and meetings helped refine and define
what the course of the research would eventually be, and as the study progressed, the
interviews became increasingly focused on data mining and gathering specific, targeted
information on the subsystems under analysis. Of note, through the interviews with the
Technical Director of PMS 400D2, a set of high-risk subsystems, their associated risk
levels, and their functional block diagrams or drawings were obtained. This information
fed directly into the logic and algorithms applied for analysis in Section 4.

3.2 Data Collection

Through the interviews with the Technical Director of PMS 400D, the three highest risk
systems were identified as the AMDR, MRG, and the MCS. All of the particulars on why
these subsystems were selected and their associated assessments are described in further
detail in Appendix A. Following the methods described by Sinha and de Weck, 2013 in
Section 2.3, the remaining data collection steps were to:

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
2
!The Technical Director of PMS [Program Management (Ships)] 400D is the lead
technical government authority for HM&E (Hull, Mechanical, and Electrical) items in the
DDG 51 program. Other systems such as Air Missile Defense Radar (AMDR) and
AEGIS are developed by other entities outside of PMS 400D. The DDG 51
government/contractor team then integrates those developed systems into the DDG 51
design.
!

! 29!
 • Assign values via expert assessment and field research for the X-vector leading to
the determination of C1.
• Gather the expert interface complexity assessment, β leading to the determination
of C2.
• Collect the subsystem functional block diagram leading to the sequential
determine of Aij, E(A), γ, and C3 as described above.

3.2.1 Component Complexity Metric, Xi

The xi metric aggregates a complexity valuation for each subsystem based upon eight
factors shown in vector X for each subsystem under analysis. It should be noted that the
X-vector as it is presented was modified and adapted for the DDG 51-oriented case study
and is thus different from what was proposed in Sinha and de Weck’s original work.

(!) (!)
!! !! = !(!"#$%#&'()"!!"#$%&'($)
(!) (!)
!! !! = !(!"#$%#&'()"!!"#"$)
(!) (!)
!! !! = ! !"#$"%&%'!!"#$
(!) (!)
(!)
!! !! = !(!"#$%&!!"!!"#$%&'!!"#$"%&"'(#) !
!
! = (!)
= (!)
; !!!!! = ! ! = !!
!! !! = !(!"#$%&!!"!!"#$"%&'() !
(!) (!)
!! !! = !(!"#)
(!) (!)
!! !! = !(!"#$%&'"!!"#$%&'(&)
(!) (!)
!! !! = !(!"#$%&'"!!"#$")
!
The!X!vector3!was!determined!through!a!combination!of!expert!assessments!and!
other!supporting!research.!!!
!
Given!the!ambiguity!and!DoD?centric!nature!of!TRL,!an!expanded!discussion!is!
required!of!TRL!and!its!effect!on!the!X6!variable.!!As!an!adaptation!to!DoD!
acquisition,!the!component!reliability!indicator,!X6!was!a!function!of!the!subsystem’s!
TRL.!!Each!factor!will!be!discussed!in!detail!later,!but!TRLs were assessed based on

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
3!Expanded!definitions!for!each!x !term!were!provided!in!Section!2.2.!
i

! 30!
information from the interview with the PMS 400D Technical Director coupled with an
integer-level corroboration per the DoD definitions for TRL. That information was
mapped to the TRL level definitions provided in Figure!7 below. !

Figure 7: TRL Level Definitions (NASA 2004) and Expanded Definitions (ASD R&E
2011)

Each subsystem’s TRL was assessed based on the capability of the manufacturer
producing that subsystem. For example, General Electric has historically produced the
highly precise MRG for the DDG 51 class. When the Flight IIA line was restarted, GE
announced that they would no longer be producing the MRG so a contract was let to
provide the MRG as GFE for the continuation of the Flight IIA class. Philadelphia Gear
Company (PGC) won the contract. When GE sent PGC the drawings for the MRG, PGC
discovered that there was missing information in the technical drawings, and that some
NRE and learning would exist until that knowledge gap between GE and PGC was

! 31!
closed. In terms of TRL, an item that was TRL 9 for GE was a TRL 5 for PGC (Hellman
2013). According to DoD definitions (ASD R&E 2011):
• TRL 5 - Fidelity of breadboard technology increases significantly. The basic
technological components are integrated with reasonably realistic supporting
elements so they can be tested in a simulated environment. Examples include
“high-fidelity” laboratory integration of components.
• TRL 6 - Representative model or prototype system, which is well beyond that of
TRL 5, is tested in a relevant environment. Represents a major step up in a
technology’s demonstrated readiness. Examples include testing a prototype in a
high-fidelity laboratory environment or in a simulated operational environment.
• TRL 9 - Actual application of the technology in its final form and under mission
conditions, such as those encountered in operational test and evaluation (OT&E).
Examples include using the system under operational mission conditions.

!"#!"#
!! =
!"#!

Where ℜ TRL ∈ 1,9 and TRLi is each subsystem’s assessed TRL based on the DoD
standard definitions. Table!3 summarizes each subsystem’s TRL and corresponding xi
value.

Table 3: Subsystem Risk and x6 Values

Subsystem TRLi x6
AEGIS AN/AEGIS 9 1.0
Air and Missile Defense Radar (AMDR) 5 1.8
Main Reduction Gear (MRGGE) 9 1.0
MRGPGC 5 1.8
Machinery Control System (MCS) 6 1.5

! 32!
 For!reference,!the!X?vector!definition!is!provided!once!more:!
!
(!)
!! = !(!"#$%#&'()"!!"#$%&'($)
(!)
!! = !(!"#$%#&'()"!!"#"$)
(!)
!! = ! !"#$"%&%'!!"#$
(!)
(!)
!! = !(!"#$%&!!"!!"#$%&'!!"#$"%&"'(#) !
! = (!)
; !!!!!!!!! = ! ! = !!
!! = !(!"#$%&!!!!!"#$"%&'()
(!)
!! = !(!"#)
(!)
!! = !(!"#$%&'"!!"#$%&'(&)
(!)
!! = !(!"#$%&'"!!"#$")
!
The!X?vector!summary!is!shown!in!Table!4.!
!
Table 4: X-Vector Summary (Normalized [0,10] Value in Parenthesis) for a possible
total Component Complexity score of 80

Component
AEGIS AMDR MCS MRG (GE) MRG (PGC)
Complexity
X1 9 9 6 10 10
X2 9 10 3 3 3
X3 7 7 3 2 2
X4 35 (10) 35 (10) 12 (3.4) 1 (0.1) 1 (0.1)
X5 35 (10) 35 (10) 31 (8.9) 27 (7.7) 27 (7.7)
X6 2.0 2.8 2.5 2.0 2.8
X7 10 3 4 6 8
X8 10 4 3 5 9
ΣXi 67 55.8 36.8 31.8 42.6
!

3.2.2 Interface Complexity Assessment, β

The Technical Director of PMS 400D also provided the assessment for interface

! 33!
complexity as described in Section 2.3. The director was asked to assess the interface
complexity between 0 and 1 with 1 being the most complex interface one would see
onboard a ship. An example of an interface that would garner a rating of 1 would be an
interface with thousands of wires of varying voltages and frequencies. On the other end
of the spectrum, an interface with a 0 rating would be quite simple, such as two plates
bolted together. “If there were multiple types of connections between two components
(say, load-transfer, material flow and control action flow), it would have a high β value
since it would be more 'complex' to achieve/design this connection compared to a simpler
load-transfer connection. For large, engineered complex systems, it appears that βij in [0,1]
is a good initial estimate.” (Sinha and de Weck 2013)

Unfortunately due to governmental classification issues, it was impossible to assess each

subsystem component’s interface at a sufficient level of detail to assign βij values for each
interace within the subsystem. Detailed information on important subsystems such as
AEGIS and AMDR are (justifiably) closely guarded secrets. Therefore, a simplying
assumption had to be made in order to obtain reasonable values for βij. It was determined
that expert assessment, via the Technical Director for PMS 400D, would suffice to
estimate the aggregated subsytem’s interface complexity β, and that overarching β value
would serve as the “rolled up” value for all the component interfaces within the
subsystem. While a realistic subsystem would most likely have interfaces of varying
complexity, the assumption is that a qualified subject matter expert could estimate the
overall average interface complexity. If the Navy chose to implement complexity-based
cost estimation as shown in this study, this simplifying assumption could be immediately
lifted to populate the βij matrix without the constraints imposed by security concerns.

Table!5 below summarizes the findings for βij for the three critical DDG 51 subsystems
ranked via the PMS 400D interview and data collection process. These values were
assumed to be the averaged value of all component interfaces contained within the
subsystem, an assumption necessary because of the lack of visibility the researcher had
into Navy subsystems because of the unclassified nature for which this study is intended.

! 34!
 Table 5: Subsystem Interface Complexity, βij (Hellman 2013)

Subsystem Interface Complexity, βij

AEGIS AN/AEGIS 0.9
Air and Missile Defense Radar (AMDR) 1.0
Main Reduction Gear (MRG) 0.3
Machinery Control System (MCS) 0.8

3.2.3 Functional Block Diagrams and Schematics

Functional block diagrams were needed to create adjacency matrix models of the system
from which the matrix energy could be derived per the methods laid out in Section 2.3.
Functional block diagrams were chosen instead of physical block diagrams due to a
subsystem’s complexity being a matter of function more than parts. A subsystem could
be comprised of a great number of parts while remaining relatively simple, while a
different subsystem could be comprised of relatively few parts while being substantially
more complex.

These diagrams led to a numerical result for C3 in the complexity equation.

AMDR and AEGIS

AMDR is slated to be the successor to the highly successful and prolific heritage AEGIS
program. On October 10th, 2013, Raytheon was awarded a $385+ million CPIF (Cost
Plus Incentive Fee) contract for the “engineering and modeling development phase
design, development, integration, test, and delivery of Air and Missile Defense S-Band
Radar (AMDR-S).” (Raytheon 2013). At the time of writing, AMDR-S and AMDR-X
(X-band capable AMDR) will reportedly fall under separate contracts, and henceforth all
references to AMDR will refer to AMDR-S.

! 35!
Because of the classified and contractually sensitive nature of AMDR, unclassified,
publically distributable data was scarce. To overcome this obstacle, the AEGIS (AMDR’s
immediate predecessor) was used as a case study to illustrate the complexity inherent in
large multi-mission shipboard radars such as AEGIS and AMDR. In terms of power and
cooling, AMDR will require substantially more of both; the current 200-ton cooling plant
will be scaled up to a 300-ton plant that will provide cooling for the AMDR plus margin
(Hellman 2013). Initially, the Flight III variant is slated to have a 14-foot AMDR that is
the maximum size the DDG 51 deckhouse can accommodate, but eventually ship designs
will be required to include the 20-foot AMDR to combat future ballistic missile threats.

A k-factor was required to quantitatively link AMDR to AEGIS in terms of complexity:

ComplexityAMDR = k ComplexityAEGIS.

To determine a reliable k factor, the key differences between AMDR and the traditional
AEGIS radars must be examined. While AMDR has an expanded mission set over the
less capable AEGIS predecessor, simply counting the missions would not necessarily
yield a reliable k factor (based on a percent increase of missions) due to the fact that
several missions are duplicated between the older AEGIS and the new AMDR.
Combining the qualitative interview assessment of increased AMDR complexity from
PMS 400D with RAND Corporation’s concept (RAND Corporation 2006) of power
density (i.e., the ratio of power generation capacity to lightship weight), an electrical
power density-based k-factor was selected. Power density yields a more reliable metric
than just power because more power does not always yield more complexity. An old
“muscle” car boasts more power than a modern luxury sedan, yet the modern luxury
sedan has orders of magnitude more complexity than the old high-powered muscle car
when one considers all the advanced technologies and software intrinsic to the new
luxury vehicle. To compare power densities, one must consider the ship as a whole, and
to facilitate the most reasonable “apples to apples” comparison, the DDG 51 Flight IIA,
as the representative for the AEGIS case, and DDG 51 Flight III, as representative of the
AMDR case, was selected. The differences between the ships in terms of length and non-
radar centric power draws are minimal; chief differences lie within the radar subsystems

! 36!
and the substantial power requirements of those radar subsystems. A k-factor was
determined for AMDR using the definition and comparison equations presented below.

!"#$%!!"#"$%&'
!!"# ! ≝ !
!"#ℎ!"ℎ!"!!"#$ℎ!

!!"#,!"#.!!! 12.0!!" 9,600!!"

!!"#$ = = ! = 1.52
!!"#,!"#.!!" 7.5!!" 9,100!!"

Based on the observed power density increase, a k-factor of 1.52 was selected as the
value estimated to yield the closest representation of the relative complexity increase in
AMDR over AEGIS, for which there is substantially more unclassified and publishable
data. Given that the k-factor functions as a coefficient adjunct to the C2C3 portion of the
complexity equation, any change in k yields a strictly linear change in net complexity so
as more complexity-oriented data becomes publicly available, kAMDR can be dialed-in to a
more accurate value. The relationships between k-factor and variations in Flight III’s
weight or power are shown in Figure!8. Variations are shown with respect to Flight III
only; values for Flight IIA were held constant.

! 37!
 Figure 8: Sensitivity Analysis for AMDR K-Factor

At the time of writing only cost estimates exist for ships that will include AMDR, namely
Flight III and beyond. The only actual costs that exist for AMDR are those in the
Research, Development, Test, and Acquisition (RDT&E) funds category so AMDR will
be used for predictive cost measures while AEGIS will be used for methodology
validation using existing costs.

Figure!9 below shows the functional block diagram for the AEGIS radar.

! 38!
 Figure 9: AEGIS Subsystem (Global Security 2011) - Unclassified

Figure!9 above was translated via functional interfaces to the symmetrical adjacency
matrix shown below in Table!6.

Correspondingly, the energy of the adjacency matrix is associated with the eigenvalues
per the equation discussed in Section 2.2:

! !!" = !! ; !!!! − !!! = 0

!!!

! 39!
It follows that:

Table 6: Adjacency Matrix, AAEGIS

!!"#$%,!!!:!" = 4.835; 4.366; 2.790; … 0 ; ! !!"#$% = ! !!"#$% = 37.33

!!!

Because there are 35 individual units (one of which has a double connection as reflected
in the adjacency matrix AAEGIS), γ 1/35 = 0.029 and C3 = γ E(AAEGIS) = 0.988.

Machinery Control System (MCS)

The MCS “provides a centralized means of monitoring and controlling the main
propulsion plant, electric plant, and auxiliary systems” (Cairns 2011) onboard the DDG
51 class. Like the AEGIS and AMDR subsystems, MCS’s complexity is derived from the
amount of interfaces, the complexity of those interfaces, and the complexity of the

! 40!
software/data contained within the system. MCS must manage over 2,000 independent
signals, which has subsequent ramifications in terms of the amount of code and wiring
involved in fielding the MCS subsystem. Further downstream, the system operators and
warfighters must manage that complexity in terms of troubleshooting errors. Finding a
loose wire in an electronics rack with 2,000 wires coming out of the back is a non-trivial
task to the operator.

MCS touches several vital systems such the power generation subsystems and electrical
distribution subsystems onboard the DDG 51, and because of MCS’s central role, efforts
have been taken to ensure its continued operation during a wartime scenario where the
host ship sustains damage. These efforts have resulted in the development of the zonal
electrical distribution system (ZEDS) that provides reliable power to vital loads via
redundancy and the compartmentalization of the electrical system into survivable zones.
As a result, MCS (as shown in Figure!10) is further complicated in reality by the
survivability requirements of the ship, but due to the unquantifiable nature of the effect of
ZEDS, MCS complexity calculations were calculated based solely on the high level block
diagram provided by NAVSEA’s PMS 400D.

! 41!
 !
Figure 10: Machinery Control System (Hellman 2013) - Unclassified

The system adjacency matrix for MCS is shown in Table!7.

! 42!
 Table 7: Binary Adjacency Matrix, AMCS
!
The figure and table show a subsystem that exhibits a high number of connections, but
the connections all lead to and from a central bus that composes the core of the MCS.

Because of the bus configuration of MCS, the adjacency matrix AMCS has only two
nonzero eigenvalues in the 31 unit length vector: σ1 = 5.48 and σ2 = 5.48.

! !!"# = ! !!"# = 10.96

!!!

!!_!"# = !! !!"# = 0.322

Recall that the C3 component of the complexity equation for AEGIS was 0.988 showing
that based on configuration alone, AEGIS was approximately three times more complex

! 43!
than MCS, which was consistent with the qualitative complexity assessments conducted
at NAVSEA and PMS 400D.

Main Reduction Gear

A reduction gear is a device used for slowing the rotational input speed to a desired
rotational output speed. In the DDG 51 class, the Main Reduction Gear (MRG) reduces
the 3600-RPM produced by the LM-2500 gas turbines to approximately 168-RPM (at full
power and 100% pitch) at the propeller shaft (Halpin 2007).

While the MRG is a highly complex and complicated subsystem, it is not complicated in
the same manner as the AEGIS, AMDR, or MCS – namely the MRG lacks the software-
intensiveness that characterize those subsystems’ core capabilities. The MRG is a
mechanical entity, but two factors have combined to make the MRG a suitable subsystem
for analysis in terms of complexity: extremely high precision requirements in terms of
position/alignment and a recent change in contractors that has effectively lowered a
heritage piece of equipment’s effective TRL as discussed in Section 2.2. Recall that the
contractor factor boosted component complexity term x6/TRL-factor by roughly 1/3 just
by changing from GE who possessed complete drawings, corporate knowledge, and a
well developed learning curve, to a new company who although possessing the technical
capability to produce MRGs did not possess those traits that led to a high effective TRL
for GE.

PMS 400D provided an unclassified schematic of DDG 106’s port MRG for analysis, and
it shown below in Figure!11.

! 44!
Figure 11: DDG 106 Propulsion Reduction Gear Arrangement, Port Gear, Isometric
View (Hellman 2013) - Unclassified

Following the same method as employed on the AEGIS and MCS subsystems, the
adjacency matrix for the MRG is shown below in Table!8.

! 45!
 Table 8: Binary Adjacency Matrix, AMRG

σMRG was comprised of five paired nonzero eigenvalues ( σ6-22 = 0):

σ!"# = ±9.29, ±1.95, ±1.00, ±1.00, ±0.98

!

! !!"# = ! !!"# = 28.44

!!!

As a result,

!!_!"# = !! !!"# = 0.836

In the following section, these values for C3 will be compared against values for C1, C2,
and Ctotal for the three highest complexity systems assessed by NAVSEA.

! 46!
4. Ship Subsystem Complexity

The subsystem data collected, formatted, and initially analyzed in Section 3 was
synthesized and underwent final analysis in Section 4 using the logic, methods, and
algorithms discussed previously.

4.1 AMDR/AEGIS, MCS, and MRG

Table!9 reflects the collection summary of data from Section 3 for synthesized and
prepared for analysis in Section 4. Mathwork’s MATLAB software was used for matrix
calculations. The code4 is attached in Appendix B.

In this thesis, complexity is determined via the overarching and governing equation for
each subsystem:
!!"#$% = !! + !! !!

Or more specifically, the expanded equation is:

! ! !

!!"#$% = ! !! + ! !!" !!" !! !

!!! !!! !!!

From the data analyzed in Section 3 using Sinha and de Weck, 2013’s methodology,
Table!9 summarizes the results. Note that the total complexity numbers report are
unitless and that the complexity of the different systems can be compared against each
other, but not necessarily against other classes of systems (such as jet engines) since
those other systems may have been assessed using different normalization factors for α
and β, among others.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
4!The author can provide a digital version of the code in the form of MATLAB’s native

m-file upon request from interested parties.!

! 47!
 Table 9: Complexity Component Results

Variable MCS MRG (GE) MRG (PGC) AEGIS AMDR

!

!! = ! !! 36.8 31.8 42.8 67.0 55.8

!!!

βi 0.8 0.3 0.3 0.9 1.0

! !

!! = !!" !!" 48.0 23.4 23.4 82.8 92.0*5

!!! !!!

C3 = γE(A) 0.353 1.053 1.053 1.098 1.327*

k-factor 1.00 1.00 1.00 1.00 1.33

Cfinal 53.8 56.4 67.2 157.91 190.1

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
5*!AMDR!values!without!objective!quantitative!substantiation!were!filled!by!the!

AEGIS!baseline!values!and!corrected!with!the!k?factors!per!the!methodology!
discussed!in!Section!3.!!The!k?factor!was!integrated!via!CFinal(AMDR!=!k*Cfinal’!

! 48!
Graphically, one can see in Figure!12 that AMDR was projected to be the most complex
system on board DDG 51 Flight III (and arguably in the U.S. Navy). AMDR’s numerical
complexity terms revealed that the relatively low component complexity C1, complexity
of the interfaces βij and C2, and the arrangement of those interfaces C3 all contribute to
the complexity of the subsystem PMS 400D had deemed “as complex as it gets.”

Figure 12: Complexity Component Breakout

For the lower complexity systems like the MRG and MCS, C1 (component complexity)
dominated, while for the higher complexity radar system, C1 played a more minor role. In
the case of AMDR, the C2C3 term dominated the net complexity of the subsystem. By

! 49!
observing the sources of complexity within each subsystem one can reasonably deduce
that the interconnections and types of connections within the subsystem can play as large
a role as the other intrinsic system properties that have hitherto predominantly been the
focus of complexity studies captured in the X-vector. Understanding the relative sources
of complexity can also help better allocate resources in the program between technical
component specialists and systems engineering.

4.2 Cost Assessment

A cost assessment was conducted to map the subsystems analyzed for complexity to their
respective costs.

While many different forms of cost exist (especially within the field of government
procurement), cost in this research was defined as either the total contract value as
announced by the winning company or the total program cost as reported by GAO reports
or publically available, Distribution A NAVSEA program office briefings.

!
Figure 13: DoD Cost Types and Relationships (DAU 2013)

! 50!
The costs used in this study are the Program Acquisition Costs outlined in the red box in
Figure!13.

4.2.1 MCS Cost Assessment

!
MCS was originally started as a NAVSEA 05 RDT&E program that was eventually
intended to undergo a parallel development effort as a separate NAVSEA 05 contract, but
program management and policy makers decided that this route included an unacceptably
high risk level in development. To drive down the risk, a new plan was created that made
MCS a separate RDT&E Line Item that was added to the lead ship contract (Fortune
2012). The first MCS was awarded as a contract to the shipyard for construction in April
of 1985 for approximately $31M ($61.1M in FY2013) for the lead unit (GAO 1990), but
further research yielded no data on the cost of subsequent units. Given the paucity of
openly available cost data, it was necessary to turn to other means of assessing a realistic
cost because most acquisitions typically witness a decrease in unit cost as the total
number of acquired units increases.

In the Department of Defense, the Learning Curve Theory is used to model “how the cost
of a good or service might vary over time…assuming [analysts] were to remove the
effects of inflation…” (DAU 2008). The effects of learning have historically been shown
to lead to a price decline as manufacturers and producers become more efficient at
producing their good or service. This assumes that the configuration of the good being
produced remains relatively stable.

The learning curves slope is the percentage value that represents the degree to which
learning is occurring as shown by the iterative decrease in unit costs over time. If the
production quantity is doubled, the slope represents as a percentage what that next
doubled values cost would be. If a notional learning curve has a slope of 80% and a unit
has an initial cost of $100, units 1, 2, 4, and 8 would cost $100, 80, 64, 51.2, respectively.
Table!10 represents typical learning curve slopes by product or trade.

! 51!
 Table 10: Typical Slope Values by Activity (DAU 2008)

!
!
Although MCS has many mechanical outputs, MCS itself is primarily a software entity,
and thus MCS should logically fall chiefly within the electronics Learning curve family.
According to Table!10, the electronics slope ranges vary by 5% within the family, and
research published by DAU stated that the variation within that family depends on a
number of factors with the most MCS-relevant being:
• Similarity between the new item and an item or items previously produced;
• Duration of time since a similar item was produced;
• Availability of material and components;

Based on the above criterion, a 95% learning curve slope was selected for use in this
analysis. This slope value was selected due to the relatively low throughput in MCS
delivery, which has historically been between 1-2 ships per year before the DDG
production shutdown. A low throughput typically corresponds to a higher learning curve
slope, which subsequently entails less cost or time reductions for the customer. This high
throughput factor is a major factor in the Korean shipyards’ cost performance, despite
sometimes building ships that are nearly identical to U.S. vessels and based upon US
MILSPEC.

! 52!
A 95% Learning slope was also chosen as the most conservative value that would reflect
the likely possibility that the people who developed this GFE equipment had prior
experience developing other MCSs for other classes of ships, such as U.S. Navy auxiliary
ships.

!
Figure 14: Machinery Control System Learning Curve Slopes. Note: while this curve
portrays learning for the MCS subsystem, the system-level curve(s) for the DDG 51 class
would be truncated at approximately unit 35 due to the fact that identical hulls are being
built at two different yards and thus undergoing two separate learning processes.

! 53!
The family of curves shown in Figure!14 reflected the substantial cost difference of one
percentage point in terms of learning curve slope distributed over 71 MCS units, from the
MCS placed onboard DDG 51 (the first of class ship in Flight I) to DDG 122 (slated last-
of-class ship for Flight IIA). Given that m is learning curve slope and x corresponds to
MCS unit number, overall cost for the MCS production run is:

!!!! $!
!"#$!"!#$ ! = 0.90 = !61.5! !, !"13
!"#$!! ! !"#!(!) !"#!(!)
!! !"#$
$!
!!! !"#$!"!#$ ! = 0.95 = 73.4 !, !"13
!"#$

The difference between just five percentage points in learning curve slope translated to
approximately $12 million over the production run for MCS.

4.2.2 MRG Cost Assessment

The MRG produced by GE originally cost $148M (FY1989, inflated to $274M in

FY2013), and was delivered as GFE for Flight I of the DDG 51 class (GAO 1990).
There were a total of 20 ships in Flight I (NAVSEA Shipbuilding Support Office 2013),
which resulted in a total of 40 MRGs produced for Flight I.

According to DDG 51 program office data, the contract for the MRG was awarded on
June 18th, 2010 to PGC for $80.2M for three ship sets with implementation starting on
DDG 113. The awarded contract was a Firm Fixed Price contract (Captain Mark
Vandroff, Program Manager, DDG 51 Shipbuilding Program 2012). As of right now, the
contract value is set at $80.2M to provide MRGs to DDG 113, 114, and 115, but options
for additional ships could bring the eventual contract value closer to $425M (Defense
Industry Daily 2012). Given that those options have not materialized at the time of
writing, $80.2M is the assumed contract value for analysis, but it is acknowledged that if
the Navy pursues the additional options, there will possibly be potential savings per unit
due to the nature of economies of scale and the effects of industrial learning. Eventually
PGC’s cost per unit will likely approach that of GE.

! 54!
4.2.3 AEGIS and AMDR Cost Assessment

!
Figure 15: RCA's 1969 Contract Initiation

The AEGIS program has a long history dating back to December 24th, 1969 when RCA
was awarded the first R&D contract to develop the AEGIS program as shown in Figure!
15. Although the program initially kicked off in the 1960s, it was not until 1986 that
RCA was able to field the system on the new DDG 51 production line. RCA along with
the AEGIS contract was sold several times in the ensuing years, but the original contract

! 55!
value for implementation on all DDGs was valued at $9.0B in FY1986, which translates
to $18.9B in FY2013 value (State New Service 1986). Since 1986, AEGIS has
undergone a series of 12 successive upgrades and modernization efforts to meet and
counter ever-evolving threats, and those efforts have come at a cost that was also factored
into this assessment. To accurately capture the cost of AEGIS, one must account for the
amount of cost that was re-invested into AEGIS for each of the successive upgrades.

Because actual and publishable cost data was unavailable for AEGIS, other software-
centric DoD systems with published cost data were analyzed for cost increases as a
percentage of original procurement cost. It was determined based on a collection of
systems from the Air Force that software-centric upgrades generally cost anywhere
between 5-10% of the original procurement cost, where more recent systems tend to have
costs at the higher end of the 5-10% range (Knaack 1988). Based on the trends published
by Knaack, those trends were extrapolated to the AEGIS upgrades and presented in Table!
11.

Table 11: AEGIS Upgrade Series Program Cost6 Estimation

Baseline U1 U2 U3 U4 U5 U6 U7 U8 U9 U10 U11 U12 Total

- 5% 5.5% 6% 6.5% 7% 7.5% 8% 8.5% 9% 9.5% 10% 10% 74.7%
$18.90B 0.95 1.04 1.13 1.23 1.32 1.42 1.51 1.61 1.70 1.80 1.89 1.89 14.13

That 1.747 factor brings the procurement total to $33.03B. The procurement quantity
was set at 71, with hulls from DDG 51 to 122; Hull 123 and beyond constitute Flight III,
and Flight III is slated to field the new AMDR.

At the time of writing, the only actual costs for AMDR are associated with RDT&E
efforts, and the acquisition is currently being split into two separate but concurrent
development efforts: one contract which was announced on October 10th, 2013 by
Raytheon for the development of the S-band portion of AMDR (AMDR-S) plus the
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
6!Program!cost!as!defined!here!subsumes!RDT&E!and!manufacturing!costs.!

! 56!
Radar Suite Controller (RSC), and the second portion for X-band radar only (AMDR-X)
will be announced at a future date. The initial AMDR-S plus RSC contract was awarded
for $1,633M for the Low Rate Initial Production (LRIP) Flight III ships (Space Daily
2013). The inclusion and integration of AMDR-X will significantly increase the cost of
the total AMDR package even beyond what is presented here, but any figures associated
with AMDR-X and the subsequent integration would only be speculation.

In this research, AMDR’s complexity and associated cost is intended to serve as a

predictive set of values, vice a set of actual values are shown on the other subsystems
that have previously been discussed in this Section.

! 57!
5. Results and Analysis

The data collected in Section 4 was analyzed via the MATLAB algorithm coded in
Appendix B. A matrix-based calculation tool was selected due to the vector array nature
of the input variables previously discussed.

5.1 Aggregated Subsystem Comparison and Analysis

The complexity data was summarized in Section 4.1, and the cost data is summarized in
Table!12. This section maps the cost and complexity data derived up to this point in an
effort to extract a quantitative relationship between those two variables using the
summary figures described in Table!9 and Table!12.

Table 12: Subsystem Cost Summary, in Millions USD

*Cost figures in $M,

MRGGE MRGPGC MCS AEGIS AMDR-S
USD
Total Program
$2747 $80 $5,208 $18,881 $1,633
Cost/Contract Value
Total Procurement
40 6 71 71 3
Quantity

Cost per Unit $6.85 $13.4 $73.35 $265.93 $544.3

Using the summarized data, each subsystem was plotted with cost versus complexity.
The results are shown in Figure!16.

The first plot correlation method studied used the power law method as proposed by
Sinha and de Weck, 2012, and for comparative purposes, an exponential curve was also
fit to the available data to determine if greater trend line fidelity could be obtained. Other
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
7!In!FY1989!the!total!value!was!$148M.!!That!figure!has!been!scaled!up!to!FY2013!

dollars,!which!takes!into!account!inflationary!effects.!

! 58!
fits such as various polynomials were also tested, but were not included here due to a
relatively low data fidelity. It should be noted that the type of curve that fits the data is
not as important as the fact that some defined quantitative relationship exists between
complexity and cost. And this relationship is non-linear i.e. cost increases superlinearly
with complexity. The aim of this study to accurately map one variable to the other is to
place a flexible and realistic decision-making tool in the hands of Navy leadership.

Figure 16: Cost versus Complexity and Cost Variance. Data points represent actual
per unit program/contract cost values obtained from company contract announcements
and GAO findings. The black trend line represents complexity-oriented cost prediction.

! 59!
Figure!16 (top) demonstrated the proposed power law curve fit to the data, which
resulted in a correlation factor8 of roughly 91.1% while the bottom figure using an
exponential fit resulted in a much higher correlation factor at 98.5% to this particular
given set of data. It is acknowledged that the limited number of data points (N = 5) in
this particular study would impede one from drawing any wide-sweeping conclusions,
but this study does demonstrate a clear trend that cost will scale with complexity in a
power-law relationship as proposed by Sinha and de Weck, 2013 or as an exponentially
increasing function as this study demonstrates. More data points would determine which
curve ultimately fits better to any given set of data, but the fact that a highly correlated
relationship exists between acquisition cost and complexity as it is quantified in this
study yields potential value in updating or augmenting Navy and DoD cost models with a
complexity-based cost estimation module. Traditional shipbuilding cost models are
mainly based on displaced volume and/or mass of the hull, however, increasingly what
goes into a ship is a bigger driver of cost than the size of the ship itself.

The small red lines in Figure!16 highlight the cost variance of the actual cost data
obtained from contract announcement and GAO reports (data points) versus the
complexity-based predictions (black line) established via the methods laid out in this
thesis.

Exponential cost growth as a function of complexity was determined to be:

$! = 23.1! !.!"#!

This means that a one-unit increase in complexity yields a 1.5% increase in acquisition
cost! Increases in complexity have occurred from myriad sources: automation, software-
intensive combat systems, and increasingly complex power systems, and this equation

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
8!Correlation (vice R2) was used to determine the power law curve fit’s fidelity whereas

R2 was used to determine the fit for the exponential data. The difference was applicability
of the two error calculation methods to the curves. Both error calculations were made via
MATLAB’s internal algorithms.
!

! 60!
tells us that those incremental boosts in complexity are responsible for an exponential
increase in cost.

It is proposed that the analysis of more subsystems would refine the coefficient values for
φ and ψ in both the power law and exponential equations while simultaneously boosting
the correlation factor. Although only four subsystems with five separate points are plotted
here, a large collection of subsystems with their respective cost data would most likely
improve the correlation number from its current value. If one were to imagine a
scatterplot of the cost and complexity of all the different systems on board the DDG 51, it
is hypothesized based on these results that the cost versus complexity trend would closely
follow either a power law or exponential curve. A model based on a larger number of
samples would serve as a powerful tool for cost analysts to either refine the current
program (PODOC) for cost estimating or possibly supplant it completely in the future.

“Rolling up” complexity, or obtaining a hierarchical quantification of the complexity

contained within any given system is a matter of assessing each subsystem’s complexity
and adding that total complexity to the next higher tier’s C1, component complexity,
term. That next higher tier’s entire complexity with the summation of the lower tiers
parts would then be added to the next level’s C1 and so on, until all the levels are
understood and quantified. While the C1 terms are a function of their components’
individual net complexity, C2 and C3 remain uniquely independent on that particular
tier’s interface complexity and topological complexity.

5.2 Sensitivity Analysis

Over the past decade, the Navy has increasingly designed ships to concepts with reduced
crew size and increased automation in an effort to reduce overall Life Cycle Cost. DDG
1000 and LCS are examples of recent high profile ship designs exemplifying a reduced
crew concept reliant upon relatively high degrees of automation.

! 61!
 In terms of subsystem architecture, an increased degree of automation implies a higher
degree of inter-connectedness within a subsystem. More parts of the ship communicate
internally with other parts of the ship with minimal reliance upon operator intervention to
do so. Although initial operational reports from LCS have indicated that automated
systems are requiring nearly as much operator management as their less technically-
advanced predecessors on other similarly-sized combatants, the large cost benefits of a
reduced crew size mean that automation will likely be improved vice being removed by
Navy decision makers.

Assuming a premise of ever-increasing automation driven by the motivation to reduce

lifecycle costs, what happens to subsystem complexity and its subsequent cost
predictions?

Relatively sparse matrices such as those shown by the bus-and-components architecture

characterized by the Machinery Control System in Figure!10 (Section 3.2.3) will most
likely have more non-zero values in the currently sparse Aij matrix as ships increasingly
move towards networked internal systems.

To more thoroughly understand the effect of inter-connectedness within the subsystem,

all factors other than the number of connections were held constant while the number of
inter-connections was allowed to vary. The number of connections in the subsystem
directly affected the energy of Aij, which subsequently affected λij, C2, and C3.

The original MCS case study had 32 connections, all of which were piecewise
connections from different subsystems to the main bus. To conduct the sensitivity
analysis, the number of interconnections was randomly boosted by a series of percentages
to study the net effect on complexity. Recall that Aij was insensitive to the type of
connections, only to the number and amount of connections involved. For example, in
terms of the complexity algorithm, it does not matter whether a hypothetical Unit A is
connected to a notional Unit B or notional Unit C, but the number and type of
connections between those units matter greatly.

! 62!
Six different notional adjacency matrices were created for MCS with the number of
connections varying between each graph. Each of the six different adjacency matrices
were created to represent notional increases in automation and were not intended to
convey any actual level of functionality in the MCS. Incremental customer-driven boosts
in automation were represented by each incremental increase in population of AMCS. The
100% connection increase is shown as an example of increased cross-connections and
automation in Table!13.

Table 13: Notional MCS Adjacency Matrix with 100% Increase in Interconnectivity

Percentage increases in connectivity were selected to determine how C3 varies with that
percentage increase. Using MATLAB and following the same logic as stipulated before,
it was discovered that C3 varies as a third order polynomial with an increase in
connectivity as shown in Figure!17. A third order polynomial fit generated a 99.9% R2

! 63!
value indicating high correlation between the data and the fitted polynomial. It should be
noted that Sinha and de Weck, 2013 predicted that C3 would vary with square root of
interconnectivity vice a third-order polynomial, but as with the original data, a larger
sample of subsystems would most likely alter the fitted equation. Regardless of whether
the fitted equation is a function of the square root, a third order polynomial, or another
related equation form, the fact that the customer may face escalating costs through an
increase in automation remains a salient and central point to this thesis.

Figure 17: C3 versus Percentage Increase in Subsystem Interconnectivity

The top graph (C2) in Figure!17 implies a strict linear relationship between an
interconnectivity increase, but the middle figure (C3) shows that above a certain point, a

! 64!
subsystem’s net C3 complexity will not vary as much as when interconnections are first
introduced into the system.

Recalling the equation that led to the calculation of C2 and C3 it can be deduced that a
linear relationship would be the inevitable product of a term comprised of multiplying the
sums:

! ! !

!!"!#$ = !! + !! !! = ! !! + ! !!" !!" !! !!"

!!! !!! !!!

However, non-linearity (in the polynomial sense) is introduced into the equation via the
matrix energy calculations, which produced the middle C3 graph shown in Figure!17.
The final result is a logical blend of the two given the multiplicative nature of the
relationship terms.

For the Navy, this means that the upfront cost of automation that is currently quite
expensive will remain so over time as each system gains evermore complexity.
Intuitively, this makes sense because a developed and interconnected system will in
effect become the new “normal” or the new baseline. Any new augmentations to the
subsystem would essentially restart you from the bottom left portion of each graph, and
the program would experience complexity costs again.

The Air Force has experienced and will experience complexity-based cost growth,
especially in terms of their increasingly prolific and highly in-demand Unmanned Aerial
Vehicle (UAV) systems. Policy makers view automation as a key way to not only drive
down costs in terms of personnel, but as a way to boost operational effectiveness by
reducing bandwidth/latency issues, improving situational awareness, and eliminating (or
mitigating) operator error through the removal of information saturation, boredom, and
sensitive judgment calls (Hansman and Weibel 2004). Figure!18 below shows DoD’s
aggressive push for UAV automation, and by recalling that cost also scales superlinearly
with complexity (driven by a boost in C3 here), one can logically deduce that the Air

! 65!
Force and DoD could have some significant and unintended cost growth in their near
future.

!
Figure 18: DOD UAV Roadmap

Based on the analysis in the C3 variable, it stands to reason that system architectures will
most likely become increasingly distributed in nature, which means that there will be
more nodes with larger numbers of connections. In an engineered system, this concept
would be represented by a cost-driven increase in systems with more parts but fewer
inter-connections meaning that interfaces would be more manageable. The obvious
tradeoff to this concept would be the coinciding boost in System Engineering,
Integration, and Test costs that would result from having a larger number of disparate
systems interacting in the aggregate.

5.3 Discussion
!
What!does!the!cost!versus!complexity!trend!really!mean?!!Recall!that!the!fitted!
equation!showed!exponential!cost!growth:!
!
$! = 23.100! !.!"#!

! 66!
If!that!relationship!was!broken!down!to!is!components!via!the!natural!log!properties!
it2is2evident2that2a2one(unit2increase2in2complexity2would2yield2a21.5%2increase2in2cost.!
!
!"($!) = !"!(23.100! !.!"#! )

!"($!) = ! ! 23.100 + 0.015!

!"($!) = 23.100 + 0.015!

One could reasonably expect that given the exponential relationship between cost and
complexity, a program or customer implementing a complexity-based cost estimation
method would set a limit on complexity units in a program, with the validated
foreknowledge that complexity is intrinsically linked to cost. From that point,
complexity-units would be a quantifiable way of managing risk and cost in the program
trade-space.

Although DDG 1000 was not analyzed in this thesis, it is reasonable to assume that the
plethora of cutting-edge (albeit complex) technology stuffed into DDG 1000 is in large
degree responsible for the staggering costs that were witnessed in the development of that
vessel. Furthermore, a complexity analysis conducted via this algorithm before
construction could possibly have given decision makers and program managers an
enhanced degree of insight into the financial path that lay before them. Program
managers could weigh the importance of considered subsystems based on complexity,
associated cost, and value to the warfighter, and it is hypothesized that some subsystems
would be devalued in importance based on that subsystem’s complexity value.

Another idea would be to implement what would essentially be a “cap and trade” style
complexity-based cost engineering approach to program management, whereby program
managers can accept greater complexity on one subsystem while trading off for a
decrease in others.

! 67!
Complexity’s exponential relationship highlights the financial danger of falling in love
with technology. The Navy is currently producing technologically unmatched vessels to
the warfighter, but this research raises the question of whether a handful of technology-
laden, multi-mission masterpieces are better than a plethora of diverse, single-mission
vessels in a cost-constrained acquisition environment.

Another question to consider is what the Navy’s function is going to be in the next war if
world events take a tragic turn for the worse? The perennial response is usually some
reference to the U.S.’s role in the Far East, where U.S. Forces are being increasingly
strategically aligned in what is known as the “Asian Pivot”. Although China is one of
America’s most-valued trading partners and both sides would see all-out war as
catastrophic turn of events for both sides, China is frequently mentioned in what-if
scenarios. In 2012, a particularly outspoken Chinese admiral in discussion of DDG 1000
frankly opined, “All it would take to sink the high-tech American ship is an armada of
explosive-laden fishing boats.” (Talmadge 2012) While the discussion of swarm-tactics
and responses are the subject of tomes of research, the comment by the Chinese admiral
is an eye-opening reminder of the technology-implementation policy contrasts in use by
different Navies around the world. Is a swarm of low-technology, single- or dual-mission
vessels superior in terms of mission-effectiveness than a handful of vessels that can
single-handedly conduct Ballistic Missile Defense, strike an Al-Qaeda or Haqqani camp
1000 miles inland, and conduct anti-submarine warfare (to name a few missions)? A
complexity-based cost assessment dictates the cost advantages of a push towards a
distributed system architecture, and it seems as though the evolutionary pressure induced
by hypothetical world threats are indicative of an additional push in the low-complexity,
high-quantity direction as well.

! 68!
6. Summary and Recommendations

As the sophistication and technology of ships increases, U.S. Navy shipbuilding must be
an effective and cost-efficient acquirer of technology-dense one-of-a-kind ships all while
meeting significant cost and schedule constraints in a fluctuating demand environment.
A drive to provide world-class technology to the U.S. Navy’s warfighters necessitates
increasingly complex ships, which further augments the non-trivial problem of providing
cost effective, on-schedule ships for the American taxpayer. The primary objective of
this study was to quantify, assess, and analyze cost predictive complexity-oriented
benchmarks in the pre-construction phase of the U.S. Navy’s ship acquisition process.
This study used commercially available software such as Mathwork’s MATLAB
software to analyze the numerical cost data and assess the fidelity of the predictive
benchmarks to the datasets. The end result was that a consideration of complexity via the
methods and algorithms established in this study supported an exponential cost versus
complexity relationship to refine the current cost estimation methods and software
currently in use in U.S. Navy shipbuilding.

With the addition of more complexity data, it would theoretically be possible to not only
present a refined exponential (or power law) cost versus complexity prediction curve, it
would also be possible to present the “rolled up” complexity for the ship in its entirety. A
validated complexity metric for the entire ship would be a powerful indicator and
comparison tool with which to compare ships of various types and classes, as well as
their respective costs. Further logical deductions could aid decision-makers in comparing
shipyards, akin to the efforts made in the benchmarking studies presented in Appendix E.

6.1 Summary

This study focused on the four main high-risk subsystems identified by PMS 400D on
board the DDG 51 Guided Missile Destroyers: the Main Reduction Gear as developed by
both General Electric and Philadelphia Gear Company, Machinery Control System,
AEGIS radar system, and the new Air and Missile Defense Radar System.

! 69!
A complexity metric was developed through 10-pronged assessment of each subsystem
based on different intrinsic properties of that subsystem and its relative relationship to
other subsystems in its technical environment.

Cost was determined via various publicly available, open-source documents including
GAO reports and Distribution A Program Office briefings.

It was found through the algorithm developed and applied in this thesis that an
exponential fit of cost data yielded a 98.5% correlation to the data. Although a sample
size of N = 5 was applied in this research, it is hypothesized based on the initial trends
that more data would confirm the exponential cost increase with complexity trend and
serve to refine the coefficient values in the final cost versus complexity exponential
equation shown in summary Figure!19.

!
Figure 19: Unit Cost vs. Complexity, Exponential Fit: $! = !". !!!.!!"#

In the future, this graph will change, as it is a snapshot in time of a system or subsystem’s
complexity and cost. Although the evidence reviewed in this study indicates a general
trend for subsystems to move up and to the right over time, there are examples where this
is not always the case. According to cost justification data produced by NSWC
Carderock, RDT&E efforts are currently underway to reduce complexity of the MCS by
developing a wireless approach that will concurrently boost the host ship’s survivability

! 70!
and reduce the overall construction costs (NSWC Carderock 2008). If and when the new
wireless MCS becomes a reality, it will shift to the left, although it remains to be seen
whether cost will correspondingly decrease.

6.2 Recommendations

The aim of this study was to examine how complexity affects cost and to leverage that
knowledge to refine the Navy’s cost estimating abilities for large-scale, complex
technical system such as those on board the various flights of the DDG 51 class. Given
the premise of that aim, the recommendations are threefold:
1. Stand up a NAVSEA 05C Integrated Product Team (IPT) to create a database of
the systems and subsystems that create the most cost vulnerability for any given
particular ship program. This database would include classified repositories of the
information that was gleaned from open-source data in this document. Each
subsystem would have an adjacency matrix verified as being correct (row-by-row)
by maintenance personnel, a corresponding interface complexity assessment as
concurred upon by the IPT, and a verified X-vector as populated by the previously
determined information and assessments.
2. For old systems with actual costs, this complexity would be evaluated for cost
versus complexity performance and would refine the overall trend-line
coefficients for the entire system. For new systems with undetermined costs, the
previous data points and their coinciding trend line would be used to make a cost
estimation that would be compared with other cost estimations as generated by
NAVSEA 05’s PODOC software tool. A record of the comparative cost
estimation performance would be tabulated.
3. Based upon the performance of this algorithm, this algorithm via a complexity-
weighted factor would either augment the conventional cost estimation methods,
or if enough fidelity existed in the long-term, complexity-based cost estimation
could completely supplant the existing methods. This would involve standing up a
cost estimation branch office that maintains, verifies, and gathers new information
regarding systems and subsystems under analysis.

! 71!
6.3 Areas for Further Study
!
This study sought to refine the Navy’s current cost models through the use of a
quantification of complexity. What would be the effect of the inclusion of other cost
estimation tools such as Compensated Gross Tonnage (CGT) whereby predictions are
made based on ship type? It is hypothesized that an algorithm comprised of a blend of
complexity and CGT could yield even further gains in the effort to shrink the ever-
present cost delta between actual and predicted system costs. Further elaboration on CGT
is included in Appendix E.

To further augment the cost prediction accuracy gains made by the inclusion of CGT-
based estimation methods, another area of study based on contract structure and the
diversity of the industrial base could prove as a useful tool for Navy and DoD policy
makers. As Dominic Alvarran pointed out in his research9 of contracts for SOW part 1 of
the Production in the Innovation Economy (PIE) study, the consequences of having only
two relatively equal market share competitors (also known as an oligopoly) with one
buyer (monopsony) has far reaching consequences characterized by the fact that free
market competition is no longer “the main driver of performance [and cost].” (Alvarran
2013) A study into the dynamics of the interplay between the actual price the Navy pays,
the price the Navy predicts they would pay via cost-estimation algorithms, contract
structure, and the number/diversity of market players could yield interesting insights and
understanding into the relatively complex DoD acquisition environment.

Roi Guinto’s 2014 research into the dynamics of Navy contract competition indicated
that when multiple sellers or vendors exist, the Navy could, on average, expect to pay 8%
less in terms of total price10 (with greater than 95% statistical confidence based on T-
tests) than when negotiating with one seller alone (Guinto 2014). With multi-year, multi-

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
9!More information on how Alvarran’s research ties into the Navy contracting process is

included in Appendix C.1.

10!Guinto noted that the 8% savings was generated due to what was termed as “shipyard

optimism” and its affect on bid competition levels, vice actual savings on the shipyard’s
part.!

! 72!
ship contract price points in the multi-billion dollar range, an 8% potential savings is
enormous, and being able to home in on an accurate price estimate would be valuable to
the taxpayer. Further cost estimation refinement studies based on complexity, CGT,
traditional SWBS groups, and competition environment levels would have the potential
to yield profound insights and open the door to fundamental change in the Navy cost
estimation process.

6.4 Closing Comments

!
Through the course of research, many valuable lessons learned and shipbuilding wisdom
has been accumulated that falls outside the relatively narrow scope of any one particular
thesis. In order to promulgate this message, a brief synopsis of these items is recorded
here at the close of my research.

• 100% of the new ship design and 100% of construction planning MUST be
complete prior to the start of construction. In the world shipbuilding market,
the US Navy is absolutely alone in the paradigm of beginning construction before
the design and plan are complete. While a concurrent design and build operation
(not to be confused with the iterative Design-Build planning technique employed
at EB and NASSCO) may slightly minimize upfront costs, the ramifications to
total overall cost and schedule due to change orders is enormous, and NASSCO’s
proprietary data has proven that the Navy simply cannot afford to continue down
this path.

For example, Daewoo Shipbuilding (NASSCO’s partner in Okpo, South Korea) will only
share a drawing for review and comments prior to the promised delivery date in order to
ensure that no changes will enter the drawing until it has been deemed complete. The
systemic nature of a ship means that small, seemingly insignificant changes in one part of
the ship can propagate throughout the entire ship touching up to 90% of the ship’s
drawings.

! 73!
Completing 100% of the new ship design and 100% of the construction planning prior to
the start of construction is the “silver bullet” to driving down cost and schedule in US
Navy shipbuilding.

The underlying pressure behind the U.S. Navy’s unique and concurrent design and build
approach is the significant political pressure applied to the program manager to meet or
exceed unreasonable schedule delivery dates (Hugel 2013).

In 2005 the GAO echoed this philosophy by stating that there was a pandemic “lack of
design maturity…required to rebuild completed areas of the ship.” (Keane 2011)

• Designing for producibility will drive down cost and improve schedule
performance. A persistent problem in shipbuilding is the tendency of the
designers to create drawings and “throw them over the wall” to the production
team. EB and NASSCO implement the Design-Build technique whereby a team
of designer and a team of manufacturing leads iterate a design for producibility,
which minimizes the problems created by the knowledge gaps of their
counterparts’ respective challenges.
A drawing MUST be signed by the production team lead as evidence of this relationship.

Designing for producibility extends to image based work packages clearly stating in
words and pictures tools required to complete the job, sequencing, and overall layout of
the construction task at hand. The roots of this effort stem from items purchased at retail
stores such as Target® or Ikea® where the end consumer constructs the final product at
home via a simple, picture-based pamphlet.

• Create, sustain, and support a culture of process improvement. The highest

performing shipyards have a robust process improvement culture with buy-in
from management and the shipyard employees. Dr. Borgschulte Germany’s
Lürssen shipyard frankly stated “It is all about process improvement”
(Borgschulte 2013), and this philosophy is echoed at NASSCO with rotating

! 74!
 “Breakfast Clubs” (early morning meetings) featuring grassroots workshop
process improvement efforts. It should be noted that not every effort or idea will
generate hundreds of thousands of dollars in savings, but the aggregation of
savings of all ideas is a substantial value to the taxpayer.

! 75!
Bibliography

Allison,!Jacob!I.!Japan2Maritime2Self(Defense2Force2(JMSDF)2guided(missle2defense2
destroyer2JDS2Atago2(DDG(177).!USN!Photo,!USN,!2011.!
!
Alvarran,!Dominic.!"Realigning!Contract!Incentives!for!the!Non?Competitive!
Environment!of!the!US!Shipbuilding!Industry."!Master's!and!Engineer's!Thesis,!
M.I.T.,!Cambridge,!MA,!2013.!
!
ASD!R&E.!Technology2Readiness2Assessment2(TRA).!Guidance,!Washington,!D.C.:!DoD,!
2011,!20.!
!
Borgschulte,!Klaus.!"Challenges!in!Specialized!Shipbuilding."!Presentation.2
!
!Cambridge,!MA:!Lurssen!Shipbuilders,!September!5,!2013.!
!
Cairns,!John!A.!"DDG!51!Class!Land!Based!Engineering!Site!(LBES)!?!The!Vision!and!
the!Value."!American2Society2of2Naval2Engineers,!2011.!
!
Captain!Mark!Vandroff,!USN.!"DDG!51!Program."!2013.!
!
Captain!Mark!Vandroff,!USN,!interview!by!US!Naval!Institute.!Program2Manager,2
DDG2512Shipbuilding2Program!www.news.usni.org,!(2012).!
!
DAU.!Cost2Terms.!2013.!!
!
https://dap.dau.mil/acquipedia/Pages/ArticleDetails.aspx?aid=419dcd24?0e78?
4279?a50a?9c122c4f0630!(accessed!February!11,!2014).!
!
DAU.!Learning2Curves.!Text,!Washington,!D.C.:!DAU,!2008.!
!
Defense!Acquisition!University.!Depot2Level2Maintenance.!2013.!
!
!https://acc.dau.mil/depot!(accessed!June!19,!2013).!
!
Defense!Industry!Daily.!Adding2Arleigh2Burkes:2H.I.I.2Steps2Forward2for2DDG(512
Restart.!June!12,!2012.!http://www.defenseindustrydaily.com/adding?arleigh?
burkes?northrop?grumman?underway?06007/!(accessed!January!7,!2014).!
!
—.!"Korea's!KDX?III!AEGIS!Destroyers."!Defense2Industry2Daily,!April!7,!2013.!
!
Dickseski,!Jerri.!"America’s!Largest!Military!Shipbuilder!Begins!Operations!as!a!New,!
Publicly!Traded!Company!under!the!Name!of!Huntington!Ingalls!Industries."!News!
Release,!HII,!2011.!

! 76!
diStefano,!Christi,!and!Gary!Davis.!Matrix2Energy.!Report,!University!of!
Massachusetts!at!Dartmouth,!Dartmouth:!CSUMS,!2009.!
!
FAS!Military!Network.!DDG(512ARLEIGH2BURKE(class.!2013.!www.fas.org!(accessed!
June!27,!2013).!
!
FMI,!and!USD(AT&L).!Globall2Shipbuilding2Industrial2Base2Benchmarking2Study.!
Benchmarking!Report,!Washington,!D.C.:!Department!of!Defense,!2005.!
!
Fortune,!Randall.!Give2and2Take:2Program2Management2Design2Engineering.!
Presentation,!Washington,!D.C.:!ASNE,!2012.!
!
Galrahn.!AMDR2Will2Bring2Very2High2Fleet2Costs.!November!17,!2011.!
!
http://www.informationdissemination.net/2011/11/amdr?will?bring?very?high?
fleet?costs.html!(accessed!October!18,!2013).!
!
GAO.!"Cost!and!Schedule!Problems!on!the!DDG?51!AEGIS!Destroyer!Program."!
Report,!Washington,!D.C.,!1990.!
!
GAO.!DEFENSE2ACQUISITIONS:2Assessments2of2Selected2Weapon2Programs.!Report,!
Washington,!DC:!GAO,!2013.!
!
GAO.!GAO2Report2GAO(12(113(05.!Report,!GAO,!Washington,!D.C.:!GAO,!2012.!
!
GAO.!Improved2Management2Practices2Could2Help2Minimize2Cost2Growth2in2Navy2
Shipbuilding2Programs.!Report,!Washington,!D.C.:!GAO,!2005.!
!
GAO.!Navy2Shipbuilding:2Cost2and2Schedule2Problems2on2the2DDG(512AEGIS2Destroyer2
Program.!Report,!Washington,!D.C.:!GAO,!1990,!60.!
!
GD!NASSCO.!MIT2Visit.!Report!Presentation,!San!Diego:!GD!NASSCO,!2013.!
!
General!Dynamics.!"2012!General!Dynamics!Annual!Report."!2012.!
!
Global!Security.!AEGIS2Combat2System2(ACS).!July!7,!2011.!!
!
http://www.globalsecurity.org/military/systems/ship/systems/aegis.htm!
(accessed!October!19,!2013).!
!
—.!World(Wide2Shipbuilding2Industry.!2010.!!
!
www.globalsecurity.org/military/world/shipbuilding.htm!(accessed!August!7,!
2013).!
!
Guinto,!Roi.!"Navy!Contract!Gaming!?!TBD."!Thesis,!MIT,!Cambridge,!MA,!2014.!

! 77!
!
Halpin,!Richard.!Propulsion2Control2the2DDG2Class:2162Years2of2Lessons2Learned.!
Washington,!DC:!www.navalengineers.org,!2007.!
!
Hansman,!and!Weibel.!Human2and2Automation2Integration2Considerations2for2UAV2
Systems.!Presentation,!Cambridge:!MIT!International!Center!for!Air!Transportation,!
2004.!
!
Harris,!Fred,!interview!by!Aaron!Dobson.!"Challenges!in!US!Navy!Shipbuilding."!
President,2GD2NASSCO2San2Diego.!San!Diego,!CA,!(August!13,!2013).!
!
Hellman,!Todd,!interview!by!Aaron!Dobson.!Technical2Director,2PMS2400D!
Washington,!DC,!(October!9,!2013).!
!
Hoffman,!Edward,!and!Matt!Kohut.!NASA's2Journey2to2Project2Management2
Excellence.!Washington,!DC:!National!Aeronautics!and!Space!Administration,!2012.!
!
Horn,!and!Johnson.!Topics2in2Matrix2Analysis.!Cambridge!Press,!1994.!
!
Hugel,!Rear!Admiral!(ret.).!"Table!Discussions!at!MIT!Shipbuilding!Seminar."!
Cambridge,!MA,!September!5,!2013.!
!
Inflation2Data.!2013.!www.inflationdata.com!(accessed!May!10,!2014).!
!
Inside2View.!2013.!www.insideview.com!(accessed!June!18,!2013).!
!
Keane,!Bob.!Reducing2Total2Ownership2Cost2(TOC):2Thinking2Outside2the2Hull.!
Presentation,!ASNE!Day!2011,!2011.!
!
Knaack,!Marcelle.!Encyclopedia2of2USAF2Aircraft2and2Missile2Systems.!Offce!of!Air!
Force!History,!1988.!
!
L?3.!L(32Maritime.!Announcment,!New!York:!L?3,!2010.!
!
Lockheed!Martin!Corporation.!"Lockheed!Martin!Awarded!Machinery!Control!
System!Computer!Program!Modernization!Contract!for!U.S.!Navy!Destroyers."!News!
Release,!Orlando,!FL,!2013.!
!
Marine!Insight.!Top2102Shipbuilding2Companies2in2the2World2in22012.!August!14,!
2012.!www.marineinsight.com!(accessed!August!7,!2013).!
!
MIT.!Production2in2the2Innovation2Economy.!2011.!http://web.mit.edu/pie.!
NASA.!NASA2Technology2Readiness2Levels.!December!16,!2004.!!
!
http://as.nasa.gov/aboutus/trl?introduction.html!(accessed!October!17,!2013).!
!

! 78!
NAVAIR.!DDG(512Machinery2Control2System2(MCS).!Flyer,!Orlando:!NAVAIR,!2013.!
!
Naval!Technology.!KDX2Class2Destroyers.!July!31,!2013.!www.naval?technology.com.!
NAVSEA!Office!of!Corporate!Communication.!DDG2512Multiyear2Procurement2
Contract2Awarded.!News!Report,!www.navy.mil,!2013.!
!
NAVSEA!Shipbuilding!Support!Office.!DDG2Class.!2013.!www.nvr.navy.mil!(accessed!
June!28,!2013).!
!
—.!Naval2Vessel2Register.!2013.!www.nvr.navy.mil!(accessed!August!7,!2013).!
NSWC!Carderock.!"Exhibit!R?2a,!RDT&E!Project!Justification."!Budget!Line!Item!
Justification,!Washington,!D.C.,!2008.!
!
Orr,!Mel.!"Republic!of!Korea!(ROK)!Navy!Destroyer!ROKS!Sejong!the!Great!(DDG!
991)."!Sea!of!Japan:!USN,!March!6,!2012.!
!
Parker,!Katrina.!"USS!Roosevelt!(DDG!80)."!Atlantic!Ocean:!USN,!July!13,!2008.!
!
RAND!Corporation.!Why2Has2the2Cost2of2Navy2Ships2Risen?!Report,!Washington,!DC:!
Arena!et!al.,!2006.!
!
Raytheon.!Raytheon2awarded2US2Navy2next2generation2Air2and2Missile2Defense2Radar2
contract.!October!10,!2013.!http://finance.yahoo.com/news/raytheon?awarded?us?
navy?next?231900443.html!(accessed!October!18,!2013).!
!
Shapiro,!Michael.!"Huntington!Ingalls!to!turn!Avondale!shipyard!into!a!builder!of!oil!
and!gas!equipment."!Daily2Press,!February!5,!2013.!
!
Sinha,!Kaushik,!and!Olivier!de!Weck.!"Structural!Complexity!Quantification!for!
Engineered!Complex!Systems!and!Implications!on!System!Architecture!and!Design."!
Proceedings2of2the2International2Design2Engineering2Technical2Conference.!Portland,!
Oregon:!American!Society!of!Mechanical!Engineers,!2013.!
!
Space!Daily.!Space2Daily.!October!15,!2013.!
!
http://www.spacedaily.com/reports/Raytheon_Awarded_US_Navy_Next_Generatio
n_Air_And_Missile_Defense_Radar_Contract_999.html!(accessed!January!7,!2014).!
!
State!New!Service.!"RCA!Fights!over!AEGIS."!The2New2York2Times,!November!30,!
1986.!
!
Talmadge,!Eric.!US2Navy2hopes2stealth2ship2answers2a2rising2China.!Yahoo!!News.!June!
4,!2012.!http://news.yahoo.com/us?navy?hopes?stealth?ship?answers?rising?china?
065329046.html!(accessed!February!12,!2014).!
!
Thomsen,!James!E.,!interview!by!PIE!Study!Team.!ASN2(RD&A)2Office!(2012).!

! 79!
!
UPI.!"RCA!Gets!Contract!for!'Aegis'."!The2News2and2Courier,!December!24,!1969:!12?
B.!
US!Government!Accountability!Office.!"GAO(13(294SP2DEFENSE2ACQUISITIONS2
Assessments2of2Selected2Weapon2Programs".!Report,!Washington,!D.C.:!GAO,!2013,!
117?8.!
!
USS2DDG2512Arleigh2Burke2Destroyer.!June!2009.!www.the?blueprints.com!(accessed!
June!27,!2013).!

! 80!
Appendices

Appendix A: Travel Summary and Interview with Technical Director, PMS 400D
May 2nd, 2013 Interviews (Cambridge, Massachusetts):
• Navy Captain and former Commanding Officer of NSWC Carderock Naval
Research Center.
• Navy Commander and submarine operations, construction, and maintenance
specialist.

August 12th to August 14th, 2013 (San Diego, California):

• Traveled to National Steel and Shipbuilding Company (NASSCO) - San Diego.
• Interviews conducted with President Fred Harris and support team.

September 16th, 2013 (Washington, D.C.):

• Visited Navy Yard to meet Technical Director for PMS 400D
• Information gathering session

October 9th, 2013 (Washington, D.C.):

• Follow up visit to Navy Yard to meet Technical Director
• Interview with TD

February 20th, 2014 (Crystal City, Virginia):

• Presented initial findings to American Society of Naval Architects at professional
society expo.
• Engaged in real-time feedback with industry experts

9 October 2013, 0900 to 1130

1100 New Jersey Avenue SE
Washington, DC 20003

The Technical Director of PMS [Program Management (Ships)] 400D is the lead
technical government authority for HM&E (Hull, Mechanical, and Electrical) items in the
DDG 51 program. Other systems such as Air Missile Defense Radar (AMDR) and
AEGIS are developed by other entities outside of PMS 400D. The DDG 51
government/contractor team then integrates those developed systems into the DDG 51
design.

! 81!
Questions for Discussion

• What systems/subsystems pose the greatest technical risk to the DDG 51

program?
o AMDR
! Significantly more complex and less understood than AEGIS
system.
! Requires more power and cooling than AEGIS. Larger cooling
plants will be implemented.
! System interface complexity assessed to be 1.0 on a 0 to 1 scale,
with 1 being the most complex. AMDR is deemed to be “as
complex as it gets”.
o Main Reduction Gear (MRG)
! Most expensive piece of HM&E
! High number of subcomponents
! New contractor will have to build new MRG.
• GE used to manufacture MRG.
• Philadelphia Gear Company (PGC) now building MRG.
• Alignment effort is critical and labor intensive.
• Quieting
! No EVM on cost performance because it is provided to the ship as
Government Furnished Equipment (GFE).
! System interface complexity assessed to be 0.3 on the 0 to 1 scale
previous mentioned.
o Mission Control System (MCS)
! High number of interfaces, wiring, and lines of code.
! Function as information “superhighway”.
! Changes to system often have unintended effects due to the high
number of systems that the MCS interfaces with.
! System interface complexity assessed to be 0.8 on the 0 to 1 scale.
• At what TRL are those systems?

! 82!
 • Why are those systems the most risky?
o Level of understanding in a new system such as AMDR
o Number of interfaces
o Number of components
o Relative contract value
• In terms of interface complexity and the understanding of those interfaces, how
would you rate interface complexity on a scale of 1 to 10? For example, the
interface complexity of one board nailed or bolted to another board would be
rated at a 1. At the other end of the spectrum, an interface composed of thousands
of wires at varying power loads, frequencies, etc. would be a 10.
o Provided above.

Data Requests

• High level functional block diagram of highest risk systems

Definition of Complexity

Definitions of complexity abound and are generally very different, but many
commonalities are observed11.
• Most complex systems contain a lot of redundancy.
• A complex system consists of many parts.
• There are many relationships/interactions among the parts.
• The complex systems can often be described with a hierarchy; redundant
components can be grouped together and considered as integrated units.

Questions and Topics of Discussion

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
11!Caprace!and!Rigo,!“A!real?time!assessment!of!the!ship!design!complexity”.!!2010!

! 83!
 1. Is vessel/system complexity currently a factor used in the creation of a
development or acquisition strategy in order to improve probability of a
successful outcome? If so, how does it factor into the process? E.g., is it a
standalone process or does it integrate into a risk assessment or risk management
strategy?
2. What indicators do you use to measure or judge complexity, if any?
3. What classes of surface ships seem to be more complex in terms of technology
systems and construction versus others?
4. What onboard ship systems, subsystems, or components stand out as having
evolved most quickly in terms of complexity? Are subsystems of differing
complexity managed differently based upon that complexity?
5. How does the novelty or newness of a technology and the organization executing
the program with that technology affect the overall assessment of complexity and
subsequent mitigation strategies? I.e., organizational factors, supply chain
complexity, national/international partners, GFE/CFE, human factors, EPA…)
6. What systems that are non-intrinsic to the ship have most increased the
“contextual” complexity of the design process?
7. In your experience, how have the program offices assessed, planned for, and
mitigated the issue of complexity? I.e., Set Based Design/Time-deferred
decision-making. What practices are most effective?
8. How have the shipyards assessed, planned for, and mitigated the issue of
complexity? What practices are most effective?
9. What aspects of the shipyard limit the yard’s ability to efficiently construct ever
increasingly complex ships?
10. What national or international industries or projects come to mind for those who
manage complexity well?

! 84!
Appendix B: MATLAB Complexity Algorithm, Source Script, and Output

Contents
1. Complexity Algorithm Inputs
2. Complexity Calculations
3. Learning Curves
4. Cost Correlation - Power Law
5. Cost Deltas - Power Law & Exponential
6. Cost Correlation - Exponential Fit
7. Algorithm Output/Display
clear all
clc
close all
format shortg

disp('--------------- Aaron Dobson, LT USN --------------------------')

disp('---- complexity.m: Algorithm to Calculate Structural Complexity ---')
disp(' In support of Naval Engineers Degree and Masters of Science in')
disp(' Engineering Management ')
disp('----------------- Submitted May 2014 ----------------------------')
disp('--------- Massachusetts Institute of Technology ------------------')
disp(' ')
disp(' ')
--------------- Aaron Dobson, LT USN --------------------------
---- complexity.m: Algorithm to Calculate Structural Complexity ---
In support of Naval Engineers Degree and Masters of Science in
Engineering Management
----------------- Submitted May 2014 ----------------------------
--------- Massachusetts Institute of Technology ------------------

Complexity Algorithm Inputs

% X1 = measure of performance tolerance

% X2 = measure of performance level
% X3 = component size indicator
% X4 = # of coupled disciplines involved
% X5 = variables involved
% X6 = TRL factor
% X7 = exisiting knowledge of operating principle
% X8 = extent of reuse/heritage indicator

% X Vector (Alpha -> C1)

X_AEGIS = [9 9 7 10 10 2.0 10 10];
X_AMDR = [9 10 7 10 10 2.8 3 4];
X_MCS = [6 3 3 3.4 8.9 2.5 6 4];
X_MRG_GE = [10 3 2 0.1 7.7 2 4 3];
X_MRG_PGC = [10 3 2 0.1 7.7 2.8 8 9];

% AMDR Complexity K-Factor

LSW3 = 9600; % Lightship weight for Flight III in Long Tons
LSW2 = 9100; % Lightship weight for Flight IIA in Long Tons
PF3 = 12; % Power for Flight III in MW

! 85!
PF2 = 7.5; % Power for Flight IIA in MW

k = (PF3/LSW3)/(PF2/LSW2); % Power Density Comparison (FIII/FIIA)

x = 5:0.05:14.95;
kP = (x./LSW3)/(PF2/LSW2);
x1 = 5000:50:14950;
kLSW = PF3*x1.^-1*(PF2/LSW2)^-1;

figure
plot(x,kP,x1/1000,kLSW)
hold on
legend('Power Variation','Weight Variation/1000')
xlabel('Variation','FontName','Times','FontSize',16);
ylabel('K-factor','FontName','Times','FontSize',16)
title('K-factor Sensitivity Analysis in Flight III','FontName',...
'Times','FontSize',18)
set(gcf,'color','w')
set(gca,'FontName','Times','FontSize',16)
H = legend('Power Variation','Weight Variation/1000');
set(H,'Location','NorthEast','FontName','Times','FontSize',14)
grid on

% Interface Complexity Assessments (Beta -> C2)

beta_AEGIS = 0.9; % beta defined as interface complexity
beta_AMDR = 1;
beta_MCS = 0.8;
beta_MRG = 0.3;

% Load EXCEL files/Adjacency Matrices (~ -> C3)

A_AEGIS = xlsread('Matrices.xlsx','AEGIS');
A_MCS = xlsread('Matrices.xlsx','MCS');
A_MRG = xlsread('Matrices.xlsx','MRG');

% Subsystem Costs:
% 1-AEGIS, 2-AMDR, 3-MCS, 4-MRG_GE, 5-MRG_PGC
cost(:,1) = [33030*0.666, 1633, 0, 274, 80.2]; % Costs in FY13 $Mil
Q(:,1) = [71, 3, 71, 40, 6]; % Quantity or units

% Learning curve SLOPE percentage for MCS

LC = 0.95; % Set value between [0.90, 0.95]

Complexity Calculations
% C1 Calculations

X = [X_AEGIS; X_AMDR; X_MCS; X_MRG_GE; X_MRG_PGC];

C1_AEGIS = sum(X_AEGIS);
C1_AMDR = sum(X_AMDR);
C1_MCS = sum(X_MCS);
C1_MRG_GE = sum(X_MRG_GE);
C1_MRG_PGC = sum(X_MRG_PGC);

% C2 Calculations

C2_AEGIS = sum(sum(beta_AEGIS*A_AEGIS));
C2_AMDR = sum(sum(beta_AMDR*A_AEGIS));
C2_MCS = sum(sum(beta_MCS*A_MCS));
C2_MRG = sum(sum(beta_MRG*A_MRG));

% C3 Calculations

E_AEGIS = sum(svd(A_AEGIS));
gamma_AEGIS = 1/length(A_AEGIS);

! 86!
C3_AEGIS = gamma_AEGIS*E_AEGIS;

E_MCS = sum(svd(A_MCS));
gamma_MCS = 1/length(A_MCS);
C3_MCS = gamma_MCS*E_MCS;

E_MRG = sum(svd(A_MRG));
gamma_MRG = 1/length(A_MRG);
C3_MRG = gamma_MRG*E_MRG;

% C = C1 + C2*C3 (Net Complexity Calculation)

C_AEGIS = C1_AEGIS + C2_AEGIS*C3_AEGIS;

C_AMDR_k = C1_AMDR + k*C2_AMDR*C3_AEGIS;
C_AMDR = C1_AMDR + C2_AMDR*C3_AEGIS;
C_MCS = C1_MCS + C2_MCS*C3_MCS;
C_MRG_GE = C1_MRG_GE + C2_MRG*C3_MRG;
C_MRG_PGC = C1_MRG_PGC + C2_MRG*C3_MRG;

% AMDR Calculations
C3_AMDR = zeros(101,1); C_AMDR1 = C3_AMDR; k1 = C3_AMDR;
for i = 1:101
k1(i) = 0.99+i/100;
eig_AMDR = eig(k1(i)*A_AEGIS);
E_AMDR = sum(abs(eig_AMDR));
C3_AMDR(i) = gamma_AEGIS*E_AMDR;
C_AMDR1(i) = C1_AMDR + C2_AMDR*C3_AMDR(i);
y1(i) = C_AMDR;
end

% Consolidated matrix with complexity sources in subsystems

C_AEGISx = [C1_AEGIS C2_AEGIS C3_AEGIS]; C(1,:) = C_AEGISx;

C_AMDRx = [C1_AMDR C2_AMDR C3_AMDR(2)*k]; C(2,:) = C_AMDRx;
C_MCSx = [C1_MCS C2_MCS C3_MCS]; C(3,:) = C_MCSx;
C_MRG_GEx = [C1_MRG_GE C2_MRG C3_MRG]; C(4,:) = C_MRG_GEx;
C_MRG_PGCx = [C1_MRG_PGC C2_MRG C3_MRG]; C(5,:) = C_MRG_PGCx;
Learning Curves
clear i j
SWslope = (90:95)/100;
cost0 = 61.1;
units = 1:71;

cost1 = zeros(6,71); costMCSx = zeros(1,6); costMCS = zeros(1,71);

for i = 1:length(SWslope)
for j = 1:length(units)
cost1(i,j) = cost0*j^log(SWslope(i))/log(2);
end
costMCSx(i) = trapz(cost1(i,:));

if SWslope(i) == LC
costMCS = cost1(i,:);
cost(3) = trapz(costMCS);
end
end
Cost Correlation - Power Law
% Aggregated Matrix

idx1 = 1:5; % 1-AEGIS, 2-AMDR, 3-MCS, 4-MRG_GE, 5-MRG_PGC

Cfinal = [C_AEGIS C_AMDR_k C_MCS C_MRG_GE C_MRG_PGC];
C(:,4) = Cfinal'; C(:,5) = idx1'; C(:,6) = cost; C(:,7) = Q;
[~,index] = sort(C(:,4),'ascend');
CF = C(index,:);

! 87!
Cx = [CF(:,1) CF(:,2).*CF(:,3)];

CQ = CF(:,6)./CF(:,7); % Cost/Quantity (Sorted)

comp = CF(:,4); % Complexity (sorted)

figure
% Finds coefficient and exponent in power law
%[slope, intercept] = logfit(comp,CQ,'loglog');
[slope, intercept] = logfit(comp,CQ,'loglog');

psi = slope;
phi = 10^intercept;

x = 0:1:max(comp)+200;
y = phi*x.^psi;

yPSI = zeros(length(x),length(x));
psi_i = 1.3:0.1:1.7;
for i = 1:length(psi_i)
for j = 1:length(x)
yPSI(i,j) = phi*j.^psi_i(i);
end
end

% Correlation of curve fit

ycorr = phi*CQ(1:5).^psi;
R = corrcoef(comp(1:5),ycorr);

Cost Deltas - Power Law & Exponential

clear i j
delta = zeros(1,5); xtemp = zeros(5,100); ytemp = xtemp; delta2 = delta;
for i = 1:size(CF,1)
delta(i) = CQ(i) - phi*comp(i)^psi;
XT(i,:) = comp(i)*ones(1,100);
YT(i,:) = linspace(CQ(i),phi*comp(i)^psi);
end
Cost Correlation - Exponential Fit
F = fit(comp,CQ,'exp1');
coeff = coeffvalues(F);
yEx = coeff(1)*exp(coeff(2)*x);
R2 = 0.9846;
[~,~,residuals] = regress(CQ,[ones(length(comp),1),comp],0.05);

clear i
for i = 1:size(CF,1)
delta2(i) = CQ(i) - coeff(1)*exp(comp(i)*coeff(2));
YT2(i,:) = linspace(CQ(i),coeff(1)*exp(comp(i)*coeff(2)));
end
Algorithm Output/Display

disp('C_final for AEGIS:')

disp(C_AEGIS)
disp(' ')
disp('C_final for AMDR, no K-factor:')
disp(C_AMDR)
disp(' ')
disp('C_final for AMDR, with K-factor = 1.33:')
disp(C_AMDR_k)
disp('C_final for MCS:')
disp(C_MCS)
disp('C_final for old GE MRG:')
disp(C_MRG_GE)
disp(' ')

! 88!
disp('C_final for new PGC MRG:')
disp(C_MRG_PGC)

figure
hold on
bar(Cx,0.5,'stack');
set(gca, 'XTick',1:5, 'XTickLabel',...
{'MRG (GE)' 'MRG (PGC)' 'MCS' 'AEGIS' 'AMDR*K-factor'})
grid on
H = legend('C_1, f(\Sigma[X_i])',...
'C_2*C_3, f(Interface Comp.,\beta | Energy(A))');
set(H,'Location','NorthWest','FontName','Times','FontSize',14)
set(gca,'FontName','Times','FontSize',16)
set(gcf,'color','w')
ylabel('Complexity Value','FontName','Times','FontSize',16)
title('Complexity Components','FontName','Times','FontSize',18)
hold off

% 1-AEGIS, 2-AMDR, 3-MCS, 4-MRG_GE, 5-MRG_PGC

figure
% Power Law Plot
subplot(2,1,1)
semilogx(comp(1),CQ(1),'rd',comp(2),CQ(2),'bd',comp(3),CQ(3),'gd',...
comp(4),CQ(4),'md',comp(5),CQ(5),'kd','LineWidth',3,'MarkerSize',8)
hold on
semilogx(x,y,'k-','LineWidth',2)
semilogx(XT(1,:),YT(1,:),'r-',XT(2,:),YT(2,:),'r-',XT(3,:),YT(3,:),'r-',...
XT(4,:),YT(4,:),'r-',XT(5,:),YT(5,:),'r-')
H = legend(sprintf('MCS, \\Delta = %0.2f $M',delta(1)),...
sprintf('MRG_{GE}, \\Delta = %0.2f $M',delta(2)),...
sprintf('MRG_{PGC}, \\Delta = %0.2f $M',delta(3)),...
sprintf('AEGIS, \\Delta = %0.2f $M',delta(4)),...
sprintf('AMDR, \\Delta = %0.2f $M',delta(5)));
set(H,'Location','NorthWest','FontName','Times','FontSize',14)
STR1 = sprintf('Power Law Fit: $M = \\phi C^{\\psi} = %0.3f C^{%0.3f}',...
phi,psi);
text(50,200,STR1,'FontName','Times','FontSize',16,'color','m',...
'BackgroundColor','w','EdgeColor','k')
set(gca,'FontName','Times','FontSize',16)
set(gcf,'color','w')
xlabel('log(Complexity)','FontName','Times','FontSize',16);
ylabel('Cost, $M','FontName','Times','FontSize',16);
axis([10 max(comp)+200 0 max(CQ)+25])
str = sprintf('Cost Variance: Cost/Unit vs. Complexity,\\phi x^{\\psi},
\\Delta_{avg.}= %0.3f $M; Corr. = %0.3f',mean(delta),R(1,2));
title(str,'FontName','Times','FontSize',18)
grid on
hold off

% Exponential Plot

subplot(2,1,2)
semilogx(comp(1),CQ(1),'ro',comp(2),CQ(2),'bo',comp(3),CQ(3),'go',...
comp(4),CQ(4),'mo',comp(5),CQ(5),'ko',...
'LineWidth',3,'MarkerSize',8)
hold on
semilogx(x,yEx,'k-','LineWidth',2)
semilogx(XT(1,:),YT2(1,:),'r-',XT(2,:),YT2(2,:),'r-',XT(3,:),...
YT2(3,:),'r-',XT(4,:),YT2(4,:),'r-',XT(5,:),YT2(5,:),'r-')
semilogx(residuals)
H = legend(sprintf('MCS, \\Delta = %0.2f $M',delta2(1)),...
sprintf('MRG_{GE}, \\Delta = %0.2f $M',delta2(2)),...
sprintf('MRG_{PGC}, \\Delta = %0.2f $M',delta2(3)),...

! 89!
 sprintf('AEGIS, \\Delta = %0.2f $M',delta2(4)),...
sprintf('AMDR, \\Delta = %0.2f $M',delta2(5)));
set(H,'Location','NorthWest','FontName','Times','FontSize',14)
set(gca,'FontName','Times','FontSize',16)
set(gcf,'color','w')
STR2 = sprintf('Exponential Fit: $M = %0.3f e^{%0.3f C}',...
coeff(1),coeff(2));
text(50,200,STR2,'FontName','Times','FontSize',16,'color','m',...
'BackgroundColor','w','EdgeColor','k')
xlabel('log(Complexity)','FontName','Times','FontSize',16);
ylabel('Cost, $M','FontName','Times','FontSize',16);
axis([10 max(comp)+200 0 max(CQ)+25])
str = sprintf('Cost Variance: Cost/Unit vs. Complexity,
\\phi_{exp}e^{C\\psi_{exp}}; \\Delta_{avg.} = %0.3f $M; R^2 =
%0.3f',mean(delta2),R2);
title(str,'FontName','Times','FontSize',18)
grid on
hold off

figure
hold on
plot(1,cost1(1,1),'ro','MarkerSize',8)
plot(units,cost1)
xlim([0.2 71])
title('Projected Machinery Control System Learning, DDG 51 to 122',...
'FontName','Times','FontSize',18)
grid on
H = legend('MCS Data: $61.1M','90%','91%','92%','93%','94%','95%');
set(gca,'FontName','Times','FontSize',16)
set(gcf,'color','w')
xlabel('Unit Number (Hull 1 to 71)','FontName','Times','FontSize',16)
ylabel('MCS Cost, $M (FY13)','FontName','Times','FontSize',16)
C_final for AEGIS:
157.91

C_final for AMDR, no K-factor:

156.81

C_final for AMDR, with K-factor = 1.33:

208.99

C_final for MCS:

53.762

C_final for old GE MRG:

56.445

C_final for new PGC MRG:

67.245

Published with MATLAB® R2014a!

! !

! 90!
Appendix C: U.S. Shipyards
!
Any discussion of current U.S. naval shipbuilding capabilities must first be predicated on
the understanding of the dichotomy that exists between a monopsony (one buyer) and an
oligopoly (multiple sellers). In the case of U.S. naval shipbuilding, the U.S. government
is the monopsonistic buyer/acquirer, and the two seller conglomerates in the oligopoly
are General Dynamics (GD) and Huntington Ingalls Industries (HII). The U.S. Navy’s
role as an acquiring agent, and both corporations’ holdings, portfolio, and subsidiaries
will be examined in further detail later.

Functionally, shipyards in the United States performing work related to U.S. Naval
vessels can roughly be divided into three categories: those performing only new
construction, those performing only repair, modernization, and overhaul (RMO), and
those that do both.

A key item affecting cost escalation is that U.S. shipyards must produce with an unstable
business base. For example, a shipyard that received a large contract to construct a new
class of say, 24 ships could realistically see that contract cut in half (if not more) due to
budget constraints and changes in defense needs. While this paradigm allows policy
makers greater flexibility in terms of planning and restructuring the force to fit needs,
shipyards are hesitant to invest in large scale capital infrastructure improvements if their
budget for making a return on that investment is unsure. This lack of reinvestment
infrastructure forces work to occur on dated or low-capacity/inefficient equipment.

C.1 U.S. New Construction Shipbuilders and Shipyards

GD and HII are the two dominant tier one shipbuilders for U.S. Naval vessels. These two
contractor conglomerates and their respective subsidiaries build all surface combatants,
support craft, and submarines. The relationship between the contractors and their
subsidiaries is shown in Figure!20.

! 91!
 General!Dynamics!(GD)

• Bath!Iron!Works! • Ingalls!

Huntington!Ingalls!
(BIW)!?!Bath,2ME2 Shipbuilding!?!

Industries(HII)!
• Electric!Boat!(EB)!?! Pascagoula,2MS2
Groton,2CT2 • !!Newport!News!
• National!Steel!and! Shipbuilding!!?!
!

Shipbuilding! Newport2News,2
Company! Virginia2
(NASSCO)!?!San2
Diego,2CA2

Figure 20: Tier One U.S. Shipbuilders and Their Subsidiaries

GD is a U.S.-based, multinational, Fortune 500 corporation with four main business

segments: Aerospace, Combat Systems, Information Technology, and Marine Systems.
Of GD’s four overarching business sectors, the marine systems segment is the smallest in
terms of revenue as shown in Figure!21 (General Dynamics 2012). The GD Marine
Systems portfolio consists of:
• Nuclear-powered Submarines (Virginia class and Ohio-class replacement)
• Surface Combatants (DDG51 and DDG1000)
• Auxiliary and Combat Logistics Ships (MLP and T-AKE)
• Commercial Ships (Jones Act ships)
• Design and Engineering Support
• Overhaul, Repair, and Lifecycle Support Services

! 92!
 Figure 21: GD Revenues by Shipyard and Sector (Inside View 2013)

Figure!21 reveals that the sum total of GD’s subsidiaries engaged in U.S. Naval
shipbuilding only accounts for 13.8% of GD’s annual revenues.

GD NASSCO (San Diego) is the only US Naval shipbuilder engaged in both naval new
construction and commercial new construction. Although other yards in Norfolk,
Mayport, and Gray fall under the GD NASSCO umbrella, those yards are engaged
exclusively in repair work. The relatively small 89-acre San Diego yard is the only one
that maintains all three activities (naval new construction, commercial new construction,
and repair) in roughly equal parts in terms of long-term average revenue (GD NASSCO
2013).

Huntington Ingalls Industries (HII) was divested from Northrop Grumman (NG) on
March 31st, 2011, and the two major shipyards in NG’s holdings were subjoined under
the new HII umbrella. HII was previously known as Northrop Grumman Shipbuilding
(NGSB) (Dickseski 2011). The two shipyards involved were:
• Ingalls Shipbuilding – Pascagoula, Mississippi
• Newport News Shipbuilding – Newport News, Virginia

! 93!
Avondale shipyard was also part of NGSB’s holdings until July 13, 2010 NGSB
announced they would be consolidating Navy shipbuilding to their Pascagoula shipyard.
At the time of writing in 2013, Navy shipbuilding is winding down in Avondale, and
HII’s intent is to migrate Avondale’s operations to oil and gas equipment construction
(Shapiro 2013). Although Navy shipbuilding is still occurring at Avondale at the time of
writing, all analysis going forward will focus on Ingalls Shipbuilding and Newport News
Shipbuilding.

Despite only engaging at shipbuilding at two shipyards, HII has captured just over 60%
over the U.S. naval shipbuilding market share as shown in Figure!22.

As Dominic Alvarran pointed out in his research of contracts for SOW part 1 of the
Production in the Innovation Economy (PIE) study, the consequences of having only two
relatively equal market share competitors (previously mentioned oligopoly) with one
buyer (monopsony) has far reaching consequences characterized by the fact that free
market competition is no longer “the main driver of performance.” (Alvarran 2013)

U.S.!Navy!New!Construction!
Apportionment!by!Revenue!(in!millions,!
USD)!
GD!BIW!(800)!

GD!EB!(2,900)!

GD!NASSCO!(650)!

HII!Ingalls!Shipbuilding!
(2,840)!
HII!Newport!News!
Shipbuilding!(3,940)!

Figure 22: U.S. Navy New Construction Apportionment by Revenue

! 94!
C.2 U.S. Repair, Modernization, and Overhaul (RMO) Shipyards

While any of the five major new construction yards can perform RMO, the bulk of the
U.S. Navy’s RMO operations occur at any of the four major public shipyards. As the
name implies, public shipyards are publicly owned, Navy-administered shipyards. In
America, there are currently four public shipyards:
• Norfolk Naval Shipyard (NNSY) – Norfolk, Virginia
• Pearl Harbor Naval Shipyard (PHNSY) – Pearl Harbor, Hawaii
• Puget Sound Naval Shipyard (PSNSY) – Bremerton, Washington
• Portsmouth Naval Shipyard (PNSY) – Portsmouth, New Hampshire

U.S. Navy ship maintenance is divided into three tiers, based on the size and capability of
the organization performing maintenance. These large public shipyards conduct Depot or
D-level maintenance, which is the largest and most capable of all three maintenance tiers.
The other two tiers are Intermediate or I-level maintenance and Organizational or O-level
maintenance. For the purposes of this study, only these four D-level maintenance
shipyards will be under analysis due to the larger budgets and small number of shipyards.
(Defense Acquisition University 2013)

To summarize, Table!14 shows the alignment of tier one, tier two, and public shipyards
aligned by the classes of USN ships they either build or conduct RMO for.

! 95!
Table 14: Tier One, Tier Two, and Public Shipyards vs. Ship Classes and Respective
Functional Alignment

!
! !

! 96!
Appendix D: R.O.K. Shipyards and the KDX-class
!

D.1 Republic of Korea (R.O.K.)’s Shipyards

In the post-World War II reconstruction era, Japan regained and expanded their pre-war
shipbuilding capabilities, and by the 1970s Korea and China had also entered the market
edging out the traditional European and U.S. shipbuilders as shown in Figure!23. In
2012, the US built only 0.1% of all the world’s ships above 100 CGT (Harris 2013).

Figure 23: Worldwide Shipbuilding Market Share (Harris 2013)

In the 1970s the Korean government established the shipbuilding industry as a strategic
national priority. During this time the government set up an initial shipbuilding cluster
around Busan with the associated equipment supplier companies in a conglomerate
known as chaebol, which initially had significant government ownership. Even to this
day, the Korean government is still the majority stakeholder in the Daewoo Shipbuilding
(DSME) who subsequently owns 40% of Doosan engine manufacturing. The number
one shipyard in the world Hyundai Heavy Industries (HHI), STX shipbuilding, and
Samsung shipbuilding all have some degree of government ownership. ROK, Japan, and
especially China all receive subsidies from their government. (Harris 2013)

! 97!
It is a popular misconception that Asian shipbuilding dominates the world market due to
cheap labor rates. While many Asian shipyards do have very cheap labor rates, the
throughput of the yards and the ensuing benefits from that throughput make the
difference that has edged out European and US shipbuilding.

In terms of gross tonnage produced, South Korea dominates the world shipbuilding
market. According to 2012 data compiled by Marine Insight based on data from the
Clarkson World Register, the top four shipbuilders in the world are all based in the ROK
(Marine Insight 2012):
1. Hyundai Heavy Industries – Ulsan, South Korea (93,893,700 GT)
2. Daewoo Shipbuilding – Okpo, South Korea (68,284,087 GT)
3. Samsung Heavy Industry – Geoje, South Korea (58,082,349 GT)
4. Hyundai Samho – Samho, South Korea (28,414,515 GT)

For comparison, US shipyards produced approximately 526,000 GT total in 2009 (Global

Security 2010).

D.2 R.O.K.’s KDX Program

The KDX (Korean Destroyer eXperimental) program was launched in 1998 to defend the
coastal waters around the Republic of Korea (ROK) and to transition the ROK Navy
(ROKN) to becoming a blue water capable force. As shown in Figure!24, Flight III of
the KDX class resembles the U.S.’s DDG 51 Flight IIA class in many aspects in that the
KDX-III is designed to be a “multi-purpose destroyer with full air defense, land attack,
anti-shipping, and anti-submarine capabilities. It is also being designed with the ability
to add tactical ballistic missile defense, [an] important consideration if North Korea is
your neighbor.” (Defense Industry Daily 2013)

! 98!
 Figure 24: Sejong the Great (DDH 991), lead ship of the KDX-III flight and DDG
80, a DDG 51 class Flight IIA vessel (Parker 2008) (Orr 2012)

As shown in Table!15 construction of the KDX program began exclusively with Daewoo
Heavy Industries.

! 99!
 Table 15: KDX Program Summary (Naval Technology 2013)

Flight Class Name Ships in Class Builder(s) Launch Dates

KDX-I Gwanggaeto the 3 DHI July 1998 to
Great June 2000
KDX-II Chungmugong Yi 6 DHI and HHI November 2003
Sun-Shin to September
2008
KDX-III Sejong the Great 3 DHI and HHI December 2008
to August 2012

! 100!
Appendix E: Benchmarking in Naval Shipbuilding
!
In May 2005 the Department of Defense (DoD) published a report entitled Global
Shipbuilding Industrial Base Benchmarking Study (GSIBBS) based on (FMI’s)
benchmarking techniques and assessments. Although multiple and proprietary editions
of this report exist, all information contained in this report is based on the publicly
releasable version that can be found at the link contained in the source bibliography.
(FMI and USD(AT&L) 2005).

FMI is a small firm based out of London, U.K., and they are considered among the best
in the world for benchmarking shipyards. The 2005 GSIBBS report produced in part by
FMI is one of the most substantial works to date to address the four concerns listed below
that make up the primary objectives of the report.

The GSIBBS report had four primary objectives:

1. Survey current manufacturing and business practices of selected global shipyards,
leveraging benchmarking work completed in previous studies.
2. Assess U.S. private shipyards using a standardized benchmarking system.
Provide specific site and comparative analysis of each major U.S. shipyard.
3. Compare the U.S. shipbuilding industry against leading international shipyards
and identify key opportunities for improvement.
4. Identify DoD, Navy, and industry actions, policies, and contract incentives to
implement remedies in the U.S. shipbuilding base.

FMI’s benchmarking system “was established in 1975 and has been refined through more
than 150 world-wide benchmarking surveys since. This benchmarking system is a
widely recognized method of assessing shipyard manufacturing and business practices.
The process also includes a normalized measure of shipyard productivity, accounting for
disparate ship complexity and varying customer profiles, to further evaluate the effective
implementation of manufacturing and business best practices.” (FMI and USD(AT&L)
2005). Figure!25 shows the underlying methods used to analyze and compare shipyards.

! 101!
 Figure 25: Analytical Process Underlying GSIBBS Findings

Step 1:
For each shipyard assessed, FMI and the DoD team evaluated 50 distinct elements within
the seven benchmarking groups shown below (FMI and USD(AT&L) 2005).

1. Steelwork Production
1.1. Plate Stockyard and Treatment 2. Outfit Manufacturing and Storage
1.2. Stiffener Stockyard 2.1. Pipe Shop
1.3. Plate Cutting 2.2. Machine Shop
1.4. Stiffener Cutting 2.3. Sheet Metal Working
1.5. Plate and Stiffener Forming 2.4. Electrical
1.6. Minor Assembly 2.5. General Storage and Warehousing
1.7. Sub-assembly 2.6. Storage of Large Heavy Items
1.8. Flat Unit Assembly
1.9. Curved and 3D Unit Assembly
1.10. Superstructure Unit Assembly 3. Pre-erection Activities
1.11. Outfit Steel 3.1. Module Building

! 102!
3.2. Outfit Parts Marshalling 6.2. Steelwork Production Information
3.3. Pre-erection Outfitting 6.3. Outfit Production Information
3.4. Block Assembly 6.4. Steelwork Coding System
3.5. Unit and Block Storage 6.5. Parts Listing Procedure
3.6. Materials Handling 6.6. Production Engineering
6.7. Design for Production
4. Ship Construction and Outfitting 6.8. Dimensional Accuracy and Quality
4.1. Ship Construction Control
4.2. Erection and Fairing 6.9. Lofting Methods
4.3. Welding
4.4. Onboard Services 7. Organization and Operating Systems
4.5. Staging and Access 7.1. Manpower and Organization of
4.6. Outfit Installation Work
4.7. Painting 7.2. Master Planning
7.3. Steelwork Scheduling
5. Yard Layout and Environment 7.4. Outfit Scheduling
5.1. Layout and Material Flow 7.5. Production Control
5.2. General Environment 7.6. Stores Control
7.7. Performance and Efficiency
Calculations
6. Design, Engineering, and Production 7.8. Quality Assurance
Engineering 7.9. Production Management
6.1. Ship Design Information Systems

Those elements were given scores from levels 1 to 5 corresponding to the following
descriptions (FMI and USD(AT&L) 2005).

Levels of Technology

1 - Reflects shipyard practice of the early 1960s.

2 - Technology employed in the modernized or new shipyards of the late 1960s and early

! 103!
1970s.

3 - Good shipbuilding practice of the late 1970s. Represented by the new or fully re-
developed shipyards of that time in the US, Europe, South Korea, and Japan.

4 - Typical of shipyards that have improved their technology during the 1980s and 1990s,
but not up to leading standards.

5 - State-of-the-art technology.

Although the publically distributable version does not include shipyard specific
information, results from shipyard specific assessment would follow the notional
example in Figure!26.

Figure 26: Notional Benchmarking Results

Step 2:
To establish shipyard productivity metrics, some compensation must be made for the
level of complexity of the vessels that the shipyard under assessment is producing. The
accepted standard for normalizing this complexity is the application of a Compensated

! 104!
Gross Tonnage (CGT) Factor where higher complexity vessels are assigned
correspondingly higher CGT Factors.

CGT Factors are determined by “characteristics such as: vessel specifications (combat
systems, survivability, shock, etc.), design standards, outfit density, average compartment
size, and the complexity of structural arrangements.” (FMI and USD(AT&L) 2005)

CGT = Vessel Gross Tonnage x CGT Factor

Figure!27 shows the comparison between a large, relatively less complex bulk carrier
and a smaller, denser, higher complexity frigate. The concept of CGT demonstrates that
the work content does not necessarily scale with the size or volume of a ship.

Figure 27: Comparison of Vessel Work Content by the CGT Method

FMI includes a customer factor as an indicator of the amount of influence the customer
has on the shipbuilding process as compared to a “normal commercial contract”. Further
discussion in Section 6 will highlight the reasons why this customer factor exists and
practical solutions to drive the highly influential customer factor down.

! 105!
Customer Factors for Various Shipbuilding Customers (FMI and USD(AT&L) 2005)

1.00 Normal Commercial Contract

1.06 Naval auxiliaries for [U.K.’s] Ministry of Defense (MoD) and typical export
combatants
1.12 Combatants built for MoD and demanding export customer

Total Work Content = CGT X Customer Factor

By taking into account CGT, the customer factor, and the total work content, productivity
at different shipyards can be compared.

Shipyard Productivity = Total Shipyard Man-hours Expended / Total Work Content

As Figure X shows, even in 2005 US significantly lagged the productivity data from the
major Asian shipbuilders in the 1990s, and although the report states that the US
improved from 3.1 in 2000 to 3.6 in 2005, a significant gap remains between the
aggregated international average and the US average.

Step 3:

Results can be aggregated by country to view trends. Figure!28 shows the proper right-
and-down progress. Later, it can be seen that the U.S. progressed in a similar manner
while still losing market share to their Asian counterparts.

! 106!
 Figure 28: Trends in Productivity vs. Use of Best Practice (South Korea)

The red line denoting naval new construction lags the black line denoting commercial
shipbuilding chiefly due to the higher complexity of naval shipbuilding and the increased
density of naval ships.

! 107!
 Figure 29: Productivity vs. Overall Best Practice Rating

Individually, the U.S. progressed in all areas as shown in Figure!29. The largest gains
were made in Steelwork Production (2.8 to 3.3) and Ship Construction and Outfitting (2.7
to 3.2), and it should be noted that on the 2000 studies both of these areas were
considered below average (FMI and USD(AT&L) 2005). GD’s NASSCO shipbuilding
in San Diego, California invested in a new automated stiffener cutting line and a new
panel line in 2003, and investments like these are examples of how the benchmarking
categories improved to the extent they did between the years 2000 and 2005.

! 108!
 Figure 30: US and International Benchmarking (FMI and USD(AT&L) 2005)

The GSIBBS report concludes with an assessment of the “significant factors undermining
US shipyard core productivity”. Many of themes discussed in these conclusions from
2005 are still a pervasive negative factor in 2013 as evidenced by interviews with
shipbuilding leaders in industry. The conclusions are given in Figure!31.

Figure 31: Significant Factors Undermining U.S. Shipyard Core Productivity

! 109!
The application of a numerical structural complexity method differs from the concept of
CGT described in Appendix D because this method will be applied to systems with the
greatest perceived complexity, uncertainty, and technological risk whereas CGT applies
to vessels in a more holistic manner. CGT numerically describes a vessel such as a
surface combatant as being more technologically complex than an oil tanker principally
due to the ratio of outfit weight to lightship weight; structural complexity will describe a
certain subsystem such as the AEGIS radar or hydraulics as being more structurally
complex than another subsystem such as the hull.
!
!

! 110!
Appendix F: Navy Official Security Approval for Release and Unlimited
Distribution

Subject: #239-14 Cost Prediction via Quantitative Analysis of Complexity

in U.S. Navy Shipbuilding

Date: Wed, 14 May 2014 13:51:45 -0400

The subject material proposed for public release has been reviewed and is
returned with the following response.

Statement A: Approved for Release. Distribution is unlimited.

This email serves as the official approval document. The printed materials
will be archived and maintained per Navy policy standards.

Deputy Director
NAVSEA Office of Corporate Communications
!
!
!
!

! 111!